.. • ,..

C'O

DEVELOPMENT OF THE EXTENDED KALMAN FILTEF

FOR THE ADVANCED COMPLETELY INTEGRATED
REFERENCE INSTRUMENTATION SYSTEM (CIRIS)

THESIS

Joseph K. Solomon, B.SM.E

Captain, USAF

IDTIC
SLECTED
DEPARTMENT OF THE AIR FORCE E
AIR UNIVERSITY

AIR FORCE INSTITUTE OF TECHNOLOGY

Wright-Patterson Air Force Base, Ohio

: ..89...i- 29 O0 "
AFIT/(; E/ENG/89M-8

DEVELOPMENT OF THE EXTENDED KALMAN FILTER

FOR THE ADVANCED COMPLETELY INTEGRATED
REFERENCE INSTRUMENTATION SYSTEM (CIRIS)

THESIS

Joseph K. Solomon, B.S.M.E

Captain, USAF

S DTIC
ELECTEft
2 9 8AR
9

E
Approved for public release; distrubution unlimited
 AFIT/G E/ENG/89M-8

DEVELOPMENT OF THE EXTENDED KALMAN FILTER

FOR THE ADVANCED COMPLETELY INTEGRATED

REFERENCE INSTRUMENTATION SYSTEM (CIRIS)

THESIS

Presented to the Faculty of the School of Engineering

of the Air Force Institute of Technology
Air University

In Partial Fulfillment of the

Requirements for the Degree of

Master of Science in Electrical Engineering

Joseph K. Solomon, B.S.M.E

Captain, USAF
* .. I, .

* .March 1989

.Approved for public release; distrubution unlimited

( .
 UNCLASSIFIED
SECURITY CLASSIFICATION OF THIS PAGE
Form Approved
REPORT DOCUMENTATION PAGE OMB No. 0704-0788
la REPORT SECURITY CLASSIFICATION lb. RESTRICTIVE MARKINGS
UNCLASSIFIED
2a. SECURITY CLASSIFICATION AUTHORITY 3. DISTRIBUTION/AVAILABILITY OF REPORT
Approved for Public Release;
2b. DECLASSIFICATION /DOWNGRADING SCHEDULE Distribution Unlimited
4. PERFORMING ORGANIZATION REPORT NUMBER(S) 5. MONITORING ORGANIZATION REPORT NUMBER(S)

AFIT/GE/ENG/89M-8
6a. NAME OF PERFORMING ORGANIZATION 6b. OFFICE SYMBOL 7a. NAME OF MONITORING ORGANIZATION
(Ifapplicable)
School of Engineering AFIT/ENG
6c. ADDRESS (City, State, and ZIP Code) 7b. ADDRESS (City, State, and ZIP Code)
Air Force Institute of Technology
Wright-Patterson AFB, OH 45433-6583

Ba. NAME OF FUNDING /SPONSORING Bb. OFFICE SYMBOL 9 PROCUREMENT INSTRUMENT IDENTIFICATION NUMBER
ORGANIZATION (Ifapplicable)
6585th Test Group GDN
ADDRESS (City, State. and ZIP Code)
8"c. 10. SOURCE OF FUNDING NUMBERS
PROGRAM PROJECT TASK WORK UNIT
Holloman AFB, NM 88230-5000 ELEMENT NO. NO. NO ACCESSION NO.

11. TITLE (Include Security Classification)

See Block 19
12. PERSONAL AUTHOR(S)
Joseph K. Solomon, Capt, USAF, BSME
13a. TYPE OF REPORT 13b. TIME COVERED 14. DATE OF REPORT (Year,Month, Day) 1S . PAGE COUNT
MS Thesis FROM ____TO -18 qrth 166
16. SUPPLEMENTARY NOTATION

17. COSATI CODES 18. SUBJECT TERMS (Continue on reverse ifnecessary and identify by block number)
FIELD GROUP SUB-GROUP - INERTIAL NAVIGATION SYSTEM, KALMAN FILTER,
7 07 03 COVARIANCE ANALYSIS, CIRIS, GLOBAL POSITIONING
SYSTEM . ....
19. ABSTRACT (Continue on reverse ifnecessary and identify by block number)
Title: Development of the Extended Kalman Filter for the Advanced
Completely Integrated Reference Instrumentation System (CIRIS)
Thesis Chairman: Lt Col Z. H. Lewantowicz

(Abstract on Back)

20. DISTRIBUTION/AVAILABILITY OF ABSTRACT 21. ABSTRACT SECURITY CLASSIFICATION

MUNCLASSIFIED/UNLIMITED 0 SAME AS RPT. C DTIC USERS UNCLASSIFIED
22a NAME OF RESPONSIBLE INDIVIDUAL 22b.TELEPHONE (Include Area Code) 22c. OFFICE SYMBOL
Lt Col Z. H. Lewantowicz 513-255-2024 AFIT/ENG
DD Form 1473, JUN 86 Previous editionsare obsolete. SECURITY CLASSIFICATION OF THIS PAGE

UNCLASSIFIED
Block 19 (Continued):

Abstract
The Completely Integrated Reference Instrumentation System (CIRIS) was devel-

oped by the Central Inertial Guidance Test Facility (CIGTF) at Holloman AFB, NM. The

CIRIS system is an inertial navigation system (INS) aided with line-of-sight range and

range-rate measurements from surveyed ground transponders. The information from the

measurement and INS data is combined using an extended Kalman filter to produce an

accurate estimate of the INS position and velocity errors. The accurate CIRIS aircraft

position and velocity data is used as a baseline reference to determine the performance

capabilities of proposed aircraft navigation systems. The new aircraft navigation systems

projected in the next five years will attain accuracies approaching the level of the current

CIRIS system. In order to test these systems, the accuracy of CIRIS will be increased

through the addition of aiding measurements from the Global Positioning System (GPS). -,

The purpose of this research is to develop the "truth" model that describes all known

error sources in the current CIRIS. Based on current literature and experiments at CIGTF,

a 127-state Kalman filter "truth" model is designed. This model includes the measurement

errors due to inaccuracies in the transponder surveys and measurement errors caused by

atmospheric propagation delays. This full-ordered Kalman filter is used to process the

empirical data from a representative CIRIS flight. The Kalman filter residuals, correlated

error states, and position/velocity error estimates are analyzed and compared to the current

CIRiS filter data to verify the accuracy of the "truth" model. Based on the 127-state
"truth" model, a 70-state reduced-ordered filter is derived to decrease computer processing

time with a minimum loss in performance. These models will serve as the foundation

for the design of an extended Kalman filter that uses the information from both transponder

and GPS measurements.

'II
 Preface
The current CIRIS system uses measurements from surveyed ground transponders to

aid its Kalman filter in estimating the errors in its inertial navigation system. The goal of

this research is to develop a Kalman filter "truth" model that describes all known error

sources in the current CIRIS system. The "truth" model developed in this research will

serve as the foundation for the future development of a Kalman filter that uses measurements

from both ground transponders and Global Positioning System (GPS) satellites. The

measurements from GPS satellites are required to increase the accuracy of current CIRIS

to a level that %ill allow it to serve as a reference baseline for testing future, high accuracy

aircraft navigation systems.

The primary tool used in conducting this research is the Multimode Simulation for

Optimal Filter Evaluation (MSOFE) software developed by the Avionics Laboratory at

Wright-Patterson AFB, OH. The power and capabilities of this package are proportional

to the current lack of documentation on how to use it. I am indebted to Mr. Robert Urbanic,

an engineer at the lab, for his guidance in learning the intricacies and anomalies of MSOFE.

Without his help this research would have never got off the ground.

I chose to use the AFIT VAX VMS computers for conducting my simulations and

software development because of the power and ease-of-use of these systems. The programs

I used on these computers pushed the machines to their computational limits. In doing this

many anomalies occurred and I must thank Mr. Daniel Zambon and Mr. Robert Ewing for

their time in helping me understand the limitations of these computer systems. Unlike the

other AFIT computer system managers, they always had time for my questions.
Finally, special thanks to my faculty advisor and mentor: Lt Col Lewantowicz. This

man defines the term workaholic, which was lucky for me. He could be found at all hours

of the day or night working away in his office and he always had time for a question. His

knowledge, experience, and ideas are an integral part of this thesis. There were many times

when he had me working at levels that I did not think were possible and without his incessant

ii
motivation I would have not been able to progress as far as I did. I feel that the knowledge

and experience I gained through our interactions will benefit me throughout my engineering

career.

Joseph K. Solomon

iii
 Table of Contents
Page

Preface...................................................................................1i

List of Figures ............................................................................. vi

List of Tables.............................................................................. viii

Abstract .................................................................................. ix

1. Introduction...................................................................... 1-1

Background ............................................................... 1-4

Research Objectives...................................................... 1-6
Research Approach ...................................................... 1-7
Overview ............................... .................................. 1-9

ii. Theory ........................................................................... 2-1

Reference Frames........................................................ 2-1

Earth Fixed Frame.................................................. 2-1
Navigation Frame .................................................. 2-1
Coordinate Transformations ...................................... 2-3
Inertial Navigation System.............................................. 2-5
Hardware............................................................ 2-5
Error State Mechanization Equations ........................... 2-6
Transponders.............................................................. 2-8
Range Measurement................................................ 2-8
Range-Rate Measurement ......................................... 2-10
Current CLRJS Filter Model ............................................. 2-10
Kalman Filter Equations.............................,*-*-.. *--*-*2-13
Correlated Error and Residual Characteristics........................ 2-17
Simulation Software ..................................................... 2-20
Multimode Simulation for Optimal Filter Evaluation
(MSOFE) ........................................................... 2-20
Profile Generator (PROFGEN) ................................... 2-21
MATRIXX........................................................... 2-22
Summary.................................................................. 2-22

III. INS "Truth" Model Design/Verification...................................... 3-1

Error States ............................................................... 3-1

INS Stochastic Differential Equation .................................. 3-2
Elements of the Fa Matrix ............................................. 3-4
Elements of the Q matrix................................................ 3-8
Alignment Simulation ................................................... 3-9
Aircraft Trajectory Simulations......................................... 3-12
Static Profile ........................................................ 3-12
Fighter Profile ...................................................... 3-12
Summary.................................................................. 3-14

iv
 Page

IV. INS/Transponder "Truth" Model Design/Verification ............................ 4-1

Range M easurement M odel ............................................................. 4-1
Range-Rate Measurem ent M odel .................................................... 4-4
Full-Ordered Kalm an Filter "Truth" M odel ..................................... 4-6
CIRIS Flight Sim ulation .................................................................. 4-9
Residual Analysis ............................................................................ 4-12
Full Ordered Filter Data ................................................................... 4-17
M eans and I o Values ............................................................. 4-17
M eans and 1 TValue Differences ........................................... 4-19
Correlated Measurement Error Analysis ......................................... 4-24
Transponder Survey Errors ..................................................... 4-24
Atm ospheric Errors ................................................................. 4-30
Calibration Errors .................................................................... 4-30
Summ ary .......................................................................................... 4-33
V. Reduced Order Filter Design/Sim ulation ................................................ 5-1

Selection of States ............................................................................ 5-1

Filter M odel ..................................................................................... 5-2
Comparison of Reduced to "Truth" ................................................. 5-3
Covariance Analysis Consideration ................................................. 5-5
Comparison of the Reduced and CIRIS filters ................................ 5-6
Position and Velocity Differences .......................................... 5-6
Correlated Error M odel Performnce ....................................... 5-10
Summ ary .......................................................................................... 5-14
VI. Conclusions and Recomm endations ....................................................... 6-1

Conclusions ....................................... 6-1

Recomm endations ............................................................................ 6-3
Additional CIRIS Flights ........................................................ 6-3
Correlated Errors ..................................................................... 6-3
Filter Tuning ........................................................................... 6-3
Acceleration Data .................................................................... 6-4
Higher Sampling Rate ............................................................. 6-4
Sm oothing Algorithm ............................................................. 6-4
GPS M easurem ents ................................................................. 6-5
Appendix A: A Historical Review of the CIRIS Development and Operation ...
A-1
Appendix B: Initial Covariance, Noise Strengths, and Dynamics Matrix Ele-
ments for the INS "Truth" Model ................................................ B-1
Appendix C: Initial Covariance, Noise Strengths, Dynamics Matrix, and Mea-
surement Gradient Matrix Elements for the Measurement "Truth"
M odel .......................................................................................... C-1
Appendix D: INS and Baro Altimeter CIRIS Flight Data ................................ D-1

Bibliography ........................................................................................................... bi- I

V ita .......................................................................................................................... v -I

v
 Listof..Figure
Figure Page

1-1. Physical Concept of CIRIS ....................................................................... 1-2

1-2. Physical Concept of Advanced CIRIS ..................................................... 1-3

1-3. Advanced CIRIS Indirect Kalman Filter .................................................. 1-4

2-1. W GS 84 Ellipsoidal M odel ...................................................................... 2-2

2-2. Navigation Fram e ..................................................................................... 2-3

2-3. Random Bias Shaping Filter Characteristics ............................................ 2-18

2-4. First Order M arkov Shaping Filter Characteristics .................................. 2-19

2-5. Gaussian "White" Noise ........................................................................... 2-20

3-1. RM S Errors for Alignment Simulation .................................................... 3-11

3-2. 1c Values for the Static Simulation ......................................................... 3-13

3-3. PROFGEN Fighter Profile ....................................................................... 3-15

3-4. la Values for the Fighter Simulation ....................................................... 3-16

4-1. Errors in Calculated M easurement ........................................................... 4-2

4-2. CIRIS Flight Trajectory ........................................................................... 4-10

4-3. CIRIS Abbreviated Flight Profile ............................................................. 4-11

4-4. Range Residuals (CIRIS and "TRUTH") ................................................. 4-14

4-5. Range-Rate Residuals (CIRIS and "TRUTH") ........................................ 4-16

vi
Figure Page

4-6. Position Error States ("TRUTH ") ............................................................. 4-18

4-7. Velocity Error States ("TRUTH ") ............................................................ 4-20

4-8. Position Differences ("TRUTH" - CIRIS) ............................................... 4-22

4-9. Velocity Differences ("TRUTH" - CIRIS) ............................................... 4-23

4-10. T-181 X ' and Y ' Correlated Errors .......................................................... 4-25

4-11. T-181 Z' and ATM Correlated Errors ...................................................... 4-26

4-12. T-185 X ' and Y ' Correlated Errors .......................................................... 4-28

4-13. T- 185 Z' and ATM Correlated Errors ...................................................... 4-29

4-14. T-186 X ' and Y ' Correlated Errors .......................................................... 4-31

4-15. T-186 Z' and ATM Correlated Errors ...................................................... 4-32

4-16. Range and Range-Rate Correlated Errors ................................................ 4-34

5-1. Position and Velocity Differences ("TRUTH" - REDUCED) ................. 5-4

5-2. Position Differences (REDUCED - CIRIS) ............................................. 5-7

5-3. Velocity Differences (REDUCED - CIRIS) ............................................ 5-8

5-4. Range Residual T-181 and T-163 ............................................................ 5-11

5-5. T-181/T-163 X' and Y' Correlated Errors ............................................... 5-12

5-6. T-181/r-163 Z' and ATM Correlated Errors ........................................... 5-13

vii
Table Page

2-1. W GS 84 C onstants ................................................................................... 2-2

2-2. Range Measurement Frequencies ............................................................. 2-9

3-1. E rror State Partitions ................................................................................ 3-3

4-1. Range Residual Statistical Comparison ................................................... 4-13

4-2. Range-Rate Residual Statistical Comparison ........................................... 4-15

4-3. Position Error State Differences ............................................................... 4-21

4-4. Velocity Error State Differences .............................................................. 4-24

5-1. "TRUTH"/Reduced Position/Velocity Error State Differences ............... 5-5

5-2. Reduced/CHRIS Position Error State Differences .................................... 5-9

5-3. Reduced/CIRIS Velocity Error State Differences .................................... 5-10

6-1. "TRUTH"/Reduced/CIRIS Position Error State Differences ................... 6-2

6-2. ITRUTH"/Reduced/CIRIS Velocity Error State Differences .................. 6-3

viii
AFIT/GE/ENG/89M-8

Abstract
The Completely Integrated Reference Instrumentation System (CIRIS) was developed

by the Central Inertial Guidance Test Facility (CIGTF) at Holloman AFB, NM. The CIRIS

system is an inertial navigation system (INS) aided with line-of-sight range and range-rate

measurements from surveyed ground transponders. The information from the measurement

and INS data is combined using an extended Kalman filter to produce an accurate estimate

of the INS position and velocity errors. The accurate CIRIS aircraft position and velocity

data is used as a baseline reference to determine the performance capabilities of proposed

aircraft navigation systems. The new aircraft navigation systems projected in the next five

years will attain accuracies approaching the level of the current CIRIS system. In order to

test these systems, the accuracy of CIRIS will be increased through the addition of aiding

measurements from the Global Positioning System (GPS).

The purpose of this research is to develop the "truth" model that describes all known

error sources in the current CIRIS. Based on current literature and experiments at CIGTF,

a 127-state Kalman filter "truth" model is designed. This model includes the measurement

errors due to inaccuracies in the transponder surveys and measurement errors caused by

atmospheric propagation delays. This full-ordered Kalman filter is used to process the

empirical data from a representative CIRIS flight. The Kalman filter residuals, correlated

error states, and position/velocity error estimates are analyzed and compared to the current

CIRIS filter data to verify the accuracy of the "truth" model. Based on the 127-state "truth"

model, a 70-state reduced-ordered filter is derived to decrease computer processing time

with a minimur- loss in performance. These models will serve as the foundation for the

design of an extended Kalman filter that uses the information from both transponder and

GPS measurements.

ix
 DEVELOPMENT OF THE EXTENDED KALMAN FILTER
FOR THE ADVANCED COMPLETELY INTEGRATED
REFERENCE INSTRUMENTATION SYSTEM (CIRIS)

I. Introduction
The Completely Integrated Reference Instrumentation System (CIRIS) is a term that

refers to a transponder aided Inertial Navigation System (INS) currently used by the Air

Force for testing new aircraft navigation systems. The Office of Primary Responsibility

(OPR) for this system is the Central Inertial Guidance Test Facility (CIGTF), 6585th Test

Group, Air Force Systems Command (AFSC), Holloman AFB, NM. The basic idea is

to fly the navigation system to be tested, the test article, and the CIRIS system over an

aircraft trajectory of interest through the CIRIS transponder range. The data from both

systems is recorded during flight and compared to determine how closely the test system

followed the CIRIS system. The CIRIS system is considered to be much more accurate

than the navigation systems currently being tested and therefore the CIRIS data serves as

a baseline for determining performance of current aircraft navigation systems (15).

The current CIRIS system is capable of providing the aircraft's latitude and longitude

with an accuracy of 13 feet (ft) lo; altitude with an accuracy of 40 ft 1o; north and west

velocity with an accuracy 0.1 ft/sec (fps) Ia; and vertical velocity with an accuracy of

0.4 fps 1o (the la value indicates a 68.3% probability that the position/velocity true

error lies in the interval between [m - a, m + y], where m is the estimated mean). The

high accuracy of the CIRIS system is due to the transponder measurement aiding. Figure

1-1 depicts the basic concept.

Ground based transponders are installed over a test range that can cover the entire

length and width of the United States land mass. The positions of these transponders are

surveyed and known precisely; CIGTF estimates the accuracy of these surveys at
l-I
 SE JS

TRA SP ND RS

Figure 1-1. Physical Concept of CIRIS

approximately 3-6 ft la. During testing the CIRIS system receives an accurate mea-

surement of the host aircraft's range and range-rate. The range and range-rate mea-

surements from each transponder are optimally combined with the CIRIS INS data, in an

extended Kalman filter, to produce an accurate estimate of the error in the INS data. The

INS data is corrected to produce an accurate estimate of the aircraft position and velocity.

Currently this data is much more accurate than the standard aircraft navigation systems

(15). A historical review of the development and operation of CiRIS is presented in

Appendix A.

To increase the accuracy of the CIRIS system it has been proposed that pseudorange

and delta-range measurements from the Global Positioning System (GPS) be added to the

transponder measurements to increase the accuracy of the CIRIS system data. This
1-2
concept is shown in Figure 1-2 and is similar to the transponder idea. The positions of

the satellites are known precisely and accurate pseudorange and delta-range measurements

from available GPS satellites can be provided to the CIRIS system.

- GPS
W /

/ /
. ,./ /

Figure 1-2. Physical Concept of Advanced CIRIS

The GPS range (RsAT), GPS range-rate (VsAT), transponder range (RTAts), trans-

ponder range-rate (VnAs), and baro altimeter (HB) measurements are optimally combined

with INS position (P), velocity (V), and acceleration (A) data, through the use of an

extendee Kalman filter, to estimate the error in the CIRIS INS. This filter concept is

shown in Figure 1-3. The CIRIS INS data is corrected, resulting in an accurate estimate

of aircraft position and velocity. This estimate is expected to be more accurate than that

of the current CIRIS system (15).

This thesis research effort concentrates on developing and verifying a realistic

full-ordered "truth" model that describes all known error sources in the current CIRIS

1-3
system (INS/Transponder). This "truth" model is the foundation for future research

into the design of an extended Kalmanfilter based on an INS/Transponder/GPSmodel;

initialresearch into this area is presented in Reference (20).

V V
INS
P
A

ALTDWETER

DAV

GPS V sAr
AV
EXTENDED
KALMAN FILTER
R~
TRANS Ap

Figure 1-3. Advanced CIRIS Indirect Kalman Filter

Background

The original CIRIS system is based on a reduced-order Kalman filter consisting of

14 states. This filter is based on the 9 primary error states of the Litton LN-15 INS, two

vertical channel aiding states, and 3 error states reserved for estimating errors in Doppler

velocity aiding measurements; the Doppler velocity error states were never used because

initial tests indicated these additional states provided an insignificant increase in accuracy.

The correlated transponder and atmospheric measurement error sources are modeled as

uncorrelated white noise. This filter is limited to 14 states, of which only 11 are being

used, due to computer limitations that existed during the initial filter design process. This

system is designated CIRIS I by CIGTF (18).

1-4
 In the current CIRIS system the Litton LN-15 INS has been replaced by the Litton

LN-39 INS. The main reason for this change was to increase the overall reliability of

the CIRIS system. This system is designated ClRIS H by CIGTF, and the current CIRIS

system data referred throughout this thesis is from the CIRIS H system (18).

Due to the low number of states, limitations exist in the Kalman filter model currently

being used by CIRIS. A higher ordered Kalman filter model, including the correlated

measurement errors, is required to increase the CIRIS filter's accuracy. Current state of

the art computational capabilities allow a higher ordered filter to be realistically

implemented. A new Kalman filter design based on the CIRIS system, with the Litton

LN-39 INS, and the addition of GPS range and range-rate aiding measurements, promise

to achieve a maximum increase in the accuracy and stability of the CIRIS velocity and

position data.

The majority of current research in this area has been directed towards the integration

of GPS measurements with INS systems. Of the many papers published in this area, the

paper by D. B. Cox (6:144-153) is most representative. This paper generally explores

the advantages of GPS integration with INS systems and serves to justify the idea of

integrating CIRIS with GPS. Previous AFIT research in this area was undertaken by R.

Butler and G. Rhue, who designed a GPS/INS system with a 15-state Kalman filter which

performed well (4).

The idea of aiding an INS with GPS and transponders is unique to CIRIS and there

are no known current papers published in this area. However, there has been one AFIT

research effort conducted by C. Smith in 1985 (19). He designed a 13-state Kalman filter

for the INS/GPS data integration. His filter performed well also, but with the advances

in technology since his thesis was written, there exists the ability to improve his filter.

The INS model used by Smith was designed for the Litton LN-15 INS which is now

obsolete. The advanced CIRIS Kalman filter will be based on the state of the art Litton

LN-39 standard INS and a Kalman filter of approximately 100 states. The models are
expanded to allow better Kalman filter error estimation.

1-5
 Past research has proven that the integration of an INS with GPS and/or transponders

can significantly increase the accuracy of aircraft navigation systems. This thesis research

effort builds on past developments and applies it specifically to the hardware proposed

for advanced CIRIS. Through the use of detailed system models and the latest technology

in Kalman filter development, this research ultimately provides an advanced CIRIS system

model with a significant increase in accuracy and stability over the current CIRIS.

Research Objectives
The first step in the efficient design of a reduced-order Kalman filter is the devel-

opment of a full-ordered "truth" model. This "truth" model must realistically model all

major error sources in the real world. Due to time and computational limitations, this

research effort concentrates on the development and verification of the "truth" model for

the current CIRIS system (INS/Transponders). Initial research completed on the

INS/GPS/'Transponder integrated model is presented in Reference (20). This goal is

defined by the following objectives:

1. Design and verify the full-ordered "truth" model for the Litton LN-39 INS using

a covariance analysis simulation.

2. Design a realistic measurement "truth" model for the range and range-rate

transponder measurements. This model includes atmospheric and transponder correlated

error sources.

3. Augment the INS and measurement models to form the overall Kalman filter
"truth" model.

4. Develop the software required to allow the full-ordered Kalman filter, based on

the "truth" model, to process the empirical data from a CIRIS flight tape. This CIRIS

flight tape contains the CIRIS INS and measurement data available during a test flight.

5. Process the data from a typical CIRIS test flight with the full-ordered Kalman

filter based on the "trut'"

model.
1-6
 6. Perform a detailed comparison of the position error states, velocity error states,

and selected transponder residuals from the current CIRIS filter and the full-ordered filter.

This comparison defines the increase in accuracy possible with a higher ordered filter and

serves to verify the full-ordered filter.

7. Perform a detailed analysis on the results of the correlated measurement states.

The purpose of this analysis is to determine if the correlated errors are fully observable

and contribute to filter error estimation.

8. Design a reduced-order filter (70 states) to reduce filter computation time while

maintaining the highest level of accuracy possible. The accuracy loss is quantified by

comparing the reduced-order filter data to the full-ordered filter.

Research Approach

First the INS "truth" model must be coded into the Multimode Simulation for

Optimal Filter Evaluation (MSOFE) simulation package. The 85-state LN-39 "truth"

model is defined in a Litton engineering analysis report (21). This report also defines

the INS error growth statistics using a 10-run Monte Carlo analysis for several different

flight trajectories. The 10-hour static and 2-hour fighter flight profiles are chosen for this

INS "truth" model verification. The MSOFE covariance simulations must agree in

magnitude and shape with the Litton Monte Carlo simulations for the MSOFE model to

be considered correct (21).

Using MSOFE, the covariance of the INS "truth" model's stochastic differential

equation is propagated over the 10-hour static flight profile; the stochastic differential

equation's solution mean is considered to be zero for all covariance simulations. The 1o

error growth plots for position, velocity, and tilts are compared with the applica-le Monte

Carlo la plots in the Litton engineering report (21).

Next, the INS "truth" model must be verified for a non-static flight profile. Using

the Profile Generator (PROFGEN) simulation package, the position, velocity, and

acceleration data for the Litton fighter flight profile is generated. Using the flight data

1-7
from PROFGEN and MSOFE, the covariance of the INS "truth" model's stochastic dif-

ferential equation is propagated over the 2-hour flight profile. Again, the 1y error growth

plots for position, velocity, and tilts are compared with the applicable Monte Carlo 1 Y

plots in the Litton engineering report (21).

The transponder measurement "truth" model developement follows. The correlated

errors in the range measurement are modeled as a random bias for the receiver/transmitter

calibration error, random biases for the transponder Defense Mapping Agency (DMA)

survey data errors, and a first order Markov atmospheric error. One random bias state is

required for the aircraft transponder receiver/transmitter bias, 3 states for each trans-

ponder's 3-dimensional survey error, and 1 state for each transponder's line-of-sight

atmospheric error.

The CIRIS system works with a window of 5-10 transponders which it samples at

a I Hertz (Hz) rate. Therefore, 10 sets of transponder bias/atmospheric states are required.

This results in 41 additional states. The correlated error in the range-rate measurement

is modeled with only a random bias error for the receiver/transmitter calibration error.

This requires only 1 additional state. The overall filter "truth" model consists of 127

states (85 INS and 42 measurement).

The 127-state "truth" model is verified using actual CIRIS flight data. The necessary

MSOFE modifications are made to facilitate this data processing. A direct comparison

is made between the "truth" model and the current CIRIS position and velocity error state

means. If the "truth" model filter is operating correctly, the "truth" model error state

means fall within the la bounds on the CIRIS error state means; since the current CIRIS

system accuracies have been independently verified. A comparison is also made between

selected transponder range and range-rate residuals. The error growth reflected in the

transponder atmospheric and survey error state means and covariances is analyzed to

determine if the states are fully observable and their magnitudes warrant inclusion in the

"truth" model.

1-8
 The final step is the design of a reduced-order Kalman filter based on this "truth"

model. The reduced-order filter is required in order to reduce the computation time for

a typical CIRIS flight to an acceptable level. The "truth" model states with insignificant

error contribution magnitudes are deleted until the computation time is reduced to an

estimated 8-12 hours of CPU time, for a 4.5 hour CIRIS flight data record, on a VAX

8650 computer system. A final comparison is made between the reduced-order filter,
"truth" model filter, and the current CIRIS filter. The final analysis quantifies the dif-

ferences between these models. The conclusions/recommendations are based on the

results of this analysis.

Overview
Chapter 2 defines the general theory that is used throughout this research. The

coordinate frames and transformations are defined. The operational concepts of current

CIRIS, the Litton LN-39 INS, and Cubic transponders are described. The basic Kalman

filter equations used in this research are defined. Finally, the MSOFE and PROFGEN

simulation software capabilities are explained.

Chapter 3 defines the Litton LN-39 INS "truth" model. The coding of this model

into MSOFE is described and the flight profiles are defined. The plots of the INS error

growth for the static and fighter flight profiles are shown and compared to the Litton

Monte Carlo error growth plots.

Chapter 4 defines the Cubic transponder range and range-rate measurement "truth"

model. In this chapter the measurement "truth" model is augmented to the INS "truth"

model and implemented in MSOFE. This full-ordered Kalman filter is used to process

a CIRIS flight tape. The position and velocity error states of the "truth" model and the

current CIRIS filter are compared and the differences quantified. The range and range-rate

residuals for selected transponders are analyzed to determine efficiency and validity of

the "truth" model. Finally, the correlated measurement error states means and covariances

are analyzed to determine extent of observability, magnitude of estimated mean, and

physical significance.

1-9
 Chapter 5 describes the reduced-order filter model. The states considered for
elimination are listed and their elimination justified. The final reduced-order filter model

is implemented in MSOFE and the CIRIS flight tape processed. Finally, the position and

velocity error growth plots are compared and the differences quantified for the

reduced-order filter, full-ordered filter ("truth"), and the current CIRIS filter.

Chapter 6 summarizes the results and emphasizes the important aspects of this

research effort. This chapter also recommends areas where future research should be

directed.

1-10
 II. Theory

This chapter describes the general theory that forms the basis for this research.

Concepts and equations are described at the level required for this effort. Where necessary,

cited references allow the reader to find more detailed information.

Reference Frames
Earth Fixed Frame. In this research it is important to relate global map

coordinates (latitude, longitude, and altitude) to an earth-centered, earth-fixed (ECEF)

coordinate system. Several different world geodetic systems exist to accomplish this task.

The most accurate system is the World Geodetic System 1984 (WGS 84). This system

is provided to the Department of Defense (DOD) by the Defense Mapping Agency. The

CIRIS INS is capable of providing latitude, longitude, and altitude in the WGS 84 system.

The transponder and satellite positions are also available in this system. Therefore, this

is the primary system used for this research.

The WGS 84 system is an ellipsoidal model of the earth as illustrated in Figure

2-1. The parameters shown in Figure 2-1, the ellipse eccentricity, and the ellipse flattening

constant are defined in Table 2-1 (8:3-10,3-11). The X' axis is parallel to the Bureau

International de L'Heure (BIH) which is defined as Zero Meridian (Greenwich Meridian).

The Y' axis is rotated 900 to the East along the equatorial plane. The Z' axis is parallel

to the earth's spin axis and completes the right-handed, earth-fixed, orthogonal coordinate

system (8:2-2). The surface of the ellipsoid is considered to be mean sea level (MSL).

Navigation Frame. The LN-39 INS implements a wander azimuth platform

mechanization. This platform frame is illustrated in Figure 2-2. The orthogonal, right-

handed, set of axes pointing North, West, and Up are designated N, W, and U respectively.

2-1
 Figure 2-1. WGS 84 Ellipsoidal Model

Table 2-1
WGS 84 Constants

Parameter Definition Value

X"II ECEF Coordinate Frame Axes none

(IE Angular Velocity of the Earth 7.292115E-5 rad/sec

A Semnimajor Axis 6378137 meters

(Equatorial Radius) (a = 2 meters)
B Semniminor Axis 6356752.3142 meters
__________ (Polar Radius)_____________

e First Eccentricity 0.0818191908426

f Flattening (Ellipticity) 0.00335281066474

G.Gravity Constant 32.08744 ft/sec2

2-2
 V|

NI

FiN e 22

f
This

frame thegue
from
N-2.navigation rame.Wethwadrngeizrote ,,ad

frame
Thaxsfteu aralt the NWU navigation frame Fa.
ardeied The gdtchatitde

is reporeeted byathaie Lhe geodtic longi.Thde byathervaiale (a) asndeeh

above the reference ellipsoid by the variable H. These 3 parameters define the vehicle

position in the geographically related navigation frame.

Coordinate Transformations. Coordinate transformations between frames

are required in many instances throughout this thesis. The 5 frames are designated T for

2-3
true reference frame, P for the platform frame, C for computation frame, N for the NWU

navigation frame, and E for the ECEF frame. The coordinatization frame of a vector is

as follows:

(_] T True T-Frame

[j C Computation C-Frame

[P Platform P-Frame
N Navigation N-Frame

E ECEF E-Frame

The true frame, platform frame, and computation frame refer to the same general

coordinate frame mechanization. However, the platform and computation frames are

slightly misaligned with respect to the true frame. The transformations from the true

frame to the platform and computation frames are defined as (13:13-1,13-2):

[P = [I + 0] [_ (2-1)

0 OZ - O Y,

= -Oz 0 Ox
OY - Ox
[_ C [+ 8] [T (2-2)

0 88 z - e
80 = -80 z 0 80 x
s0e: - Zex 0

The 50 and € variables are skew-symmetric matrices representing the small misalignment

angles. This concept is used by Litton in the derivation of the INS error equations.

The majority of vectors used throughout this thesis are coordinatized in the platform

frame. Therefore, to simplify notation, any vector shown without coordinatization

brackets is assumed to be coordinatized in the P frame. The coordinate transformation

matrices for the P, N, and E frames are now defined:

2-4
 NI (2-3)
cos() - sin(a) 0
C = Sin(a) cos(a) (2-4)
10 0 Oi

[N =c [LN (2-5)
[- sin(L) cos(X) sin(k) cos(L) cos(X)1
C = - sin(L) sin(X) - cos(X) cos(L) sin(X) (2-6)
cos(L) 0 sin(L) I

CN =PC C (2-7)

[- sin(L) cos(X) cos(cc) + sin(X) sin(or)

CP = - sin(L) sin(X) cos(c) - cos(X) sin(a)
cos(L) cos(c)
sin(L) cos(X) sin(a) + sin(X) cos(a) cos(L) cos(X)" (2-8)
sin(L) sin(X) sin(ct) - cos(X) cos(a) cos(L) sin(X)
- cos(L) sin(cc) sin(L)

Inertial Navigation System

Hardware. The original INS used in CIRIS was the Litton LN-15. This INS has

been replaced by CIGTF with the reliable and accurate Litton LN-39 standard INS. The

LN-39 INS is also the standard navigation unit for the USAF F-16 fighter aircraft. This

section describes the major operating characteristics of this INS.

The LN-39 is implemented in a wander-azimuth, local-level mechanization with a

4-gimbal, all-attitude platform. Vertical channel stabilization is accomplished with

baro-altimeter aiding using a 3-state filter. This system is designed by Litton to meet the

form, fit, and function characteristics and low life-cycle cost objectives of the USAF

moderate-accuracy INS as specified in ASD/ENAC 77-1. This USAF standard INS,

designated AN/ASN-141, is designed for multiple aircraft applications and worldwide

operation (10:1 -1).

2-5
 The LN-39 INS is based on the Litton P-1000C all-attitude inertial platform, G-1200

gyros, A-1000 accelerometers, LC-4516 digital computer, and a MIL-STD-1553 data bus.

This INS utilizes gyrocompass or stored heading initial alignment procedures. The user

can also select from one of 11 different earth reference spheroids. In the CIRIS operational

mode, the gyrocompass alignment and WGS-84 earth reference spheroid are used. In this

operating mode the INS achieves accuracies of 0.8 nautical miles/hour (Circular Error

Probability (CEP)), 2.5 fps 1c horizontal velocity, and 2.0 fps la vertical velocity

(10:1-2,1-4,1-14).

Error Mechanization Equations. The concept of terrestrial navigation using

inertial equipment involves the measurement of the specific force vector by an orthogonal

triad of accelerometers. The orientation relative to the earth of the accelerometer triad is

established and maintained by a stable mounting platform. This platform is instrumented

with three orthogonally mounted gyroscopes whose output is used to control the attitude

of the platform. The force vector due to gravity is analytically removed from the

acceleration measurements to obtain the acceleration of the inertial system's center of

mass, relative to inertial space. This inertial acceleration is compensated for Corriolis

accelerations and earth rate, with respect to the coordinatization of the accelerometer

reference frame (in this case the wander azimuth frame). Using appropriate initial con-

ditions, these variables are integrated to obtain the inertial velocity of the system. This

nonlinear vector differential equation for velocity has the form (13:13-1):

V = A + C - y (2-9)

A = Acceleration Measurements

C = Calculated Coriolis Accelerations

y = Modeled Gravity Vector

2-6
This terrestrial velocity, referenced in the inertial frame, is transformed to an earth-

referenced velocity. Finally, using appropriate intial conditions, the velocity vector is
integrated to obtain the system position in the geographic navigation frame. The nonlinear

position equations have the form:

L = -o (2-10)

= w sec(L) (2-11)

(x = a:Z - oN tan(L) (2-12)

H = V z (2-13)

(Ow = f(Y)

The nonlinear equations defined in Eqs (2-9) through (2-13) are combined to form the
describing nonlinear vector differential equation of the form:
X(t) = fX(t)l

In order to analyze the error characteristics of these nonlinear equations, these

equations are expanded in a Taylor series and truncated to first order. This forms the

9-state navigation system linearized error dynamics model. These linearized dynamics
are in the standard form used by the linearized Kalman filter system model (13:13-2).

The Taylor series expansion is about a nominal point. It is standard procedure in navi-

gation applications to use the corrected INS data (corrected using Kalman filter estimated

errors) as the nominal point. These quantities are defined as:

S= ls - SI. (2-14)

X.s - (2-15)

=Hs - S1-I (2-16)

. = OCs - S (2-17)

The "hat" designated delta variables are the repective errors estimated by the Kalman

filter and the "hat" designated variables are the "best" estimate of the respective nominal

or true point.

2-7
 The engineers at Litton augmented the correlated error sources specific to the LN-39

INS to these linearized equations to the form the 85 state "truth" model defined in Ref-

erence (21). The details of this derivation are not presented.

Transpondei s
The transponder measurement system was developed by the Cubic Corporation (7).

This system provides accurate line-of-sight (slant) range and range-rate measurements.

The system is designated by Cubic as the Range/Range-Rate Subsystem (RRS). This

section describes its major operating characteristics and accuracies.

The RRS consists of an airborne interrogator (receiver/transmitter) and ground-based

transponders (receiver/transmitter). The interrogator transmits/receives carriers in the

frequency bandwidth 2400-2500 MHZ. The interrogator operation is controlled by the

CIRIS airborne computer. The specific transponders to be used during a flight are spe-

cified to the CIRIS computer before the flight. The CIRIS computer normally cycles

through a window of 5-10 transponders at a 1 Hz rate. The transponder window is

changed, as the flight progresses, by the CIRIS computer to optimize the aircraft to

transponder distance.

The CIRIS computer interrogates the transponder first to verify the transponder is

fully operational. Once the transponder status is verified as good, the CIRIS computer

directs the RRS to lock-on to the transponder and activate the range/range-rate mea-

surement loops (7:1).

Range Measurement. A frequency synthesizer in the interrogator generates

the range modulation and transmits it on the primary carrier to the transponder. It is

re-transmitted by the transponder back to the interrogator receiver. The round-trip phase

delay is measured and converted to range.

2-8
 The range measurement is based upon the principle that sine wave modulation on

a radio frequency (RF) carrier, propagated through free space. It undergoes a phase shift

directly proportional to the distance traveled. This phase shift is independent of carrier

frequency and can be measured by means of an electronic phase-meter. For nonambiguous

measurements, the modulation wavelength must be long enough so that the largest

measurable delay (one-half wavelength) corresponds to a large range increment. However,

for good resolution, a short wavelength is needed so the smallest measurable phase delay

corresponds to a sm. 1I increment of range. In the RRS, four harmonically related fre-

quencies (tones) are used. The four frequencies are required to allow the system to

measure the maximum range required by CIRIS with a resolution of 1 ft. The frequencies

and their characteristics are shown in Table 2-2.

The actual transmission method for these frequencies and the calculation of the

range magnitude is beyond the scope of this section, but further detail can be found in

Reference (7). Range is measured from 200 ft to 200 miles with an overall accuracy

of 3 ft Io (7:1,43,44).

Table 2-2
Range Measurement Frequencies

Designation Frequency Half Wavelength Resolution

(KHZ) (ft) (ft)

Fine 240.053 2,048 1

Intermediate 30.007 16,384 8

Coarse 3.751 131,072 64

Very Coarse 0.469 1,048,576 512

2-9
 Range-Rate Measurement. An oscillator in the interrogator generates the

range-rate reference frequency and transmits it on the primary carrier to the transponder.

It is retransmitted by the transponder back to the interrogator receiver. The round-trip

doppler shift is measured and converted to range-rate,

The actual transmission method for this frequency and the calculation of the

range-rate magnitude is beyond the scope of this section, but further detail can be found

in reference (7). Range-rate is measured from -5000 fps to +5000 fps with range

accelerations from -1000 fps2 to +1000 fps2 . The overall accuracy is 0.03 fps la (7:1,43).

Current CIRIS Filter Model

This section describes the major characteristics of the current CIRIS filter model.

The 11 -state filter model is described by the stochastic error differential equation shown

in Eq (2-18).

SX(t) = F(t) X(r) + W(t) (2-18)

E{W(t)} = 0

E { W(t) W(t +,[) } = Q(t) 8(T)

The error state vector is defined as:

Latitude Error
8k Longitude Error
SH Altitude Error
ON North Tilt
OF East Tilt
X = z = Vertical Tilt
8
VN North Velocity Error
8VE East Velocity Error
8'Vz Vertical Velocity Error
8iB Baro Altimeter Error
._8A Vertical Acceleration Errod

2-10
The measurement equation is defined in Eq (2-19). The index k=l for the range mea-

surement and k=2 for the delta-range measurement. In this research these two mea-

surements are considered to completly uncorrelated in time. Therfore the cross correlation

elements in the defining 2x2 noise stength matrix are zero and the scalar R values defined

in Eq (2-19) represent the diagonal elements of this matrix. The delta-range measurement

is equal to the change in the range measurement (ft) over the time of the measurement

cycle. The time of the measurement cycle normally ranges from 100 to 900 milliseconds.

84(t) = Hk(tI) X_Q) + Vk(t) (2-19)

E{Vk(t,)} = 0

E { Vk(r) Vk(tj) I = R(, t)=,tj

10 tj # t

The actual structure of these equations are beyund the scope of this section, but further

information can be found in Reference (14).

It is important to note how the time correlated INS and measurement errors are taken

into account in the filter error estimates. The current CIRIS model is limited to the

standard INS position, velocity, tilt, and vertical channel aiding error states. The correlated

INS errors such as gyro drift rate, gyro bias, accelerometer bias, etc., are not modeled.

The correlated measurement errors such as transponder survey errors, atmospheric errors,

and transponder interrogator calibration errors are not modeled. However, these errors

are compensated in the model by using time varying noise strengths (R and Q). The

system noise strength (Q) is varied to compensate for INS gyro and accelerometer cor-

related error-. The measurement noise strength (R) is varied to compensate for trans-

ponder survey, atmospheric errors and transponder interrogator calibration errors. The

filter uses a complex algorithm based on different parameters such as INS velocity,

accelerations, and range to the transponder. The details of this algorithm are described

in Reference (14).

2-11
 An additional open loop deterministic correction is made to the line-of-sight range

measurement data before it is processed by the filter. This correction is shown in Eq

(2-20).

RM = RM - e (2-20)

RM = Corrected Range Measurement

RM = Raw Range Measurement

eR = Error Due To Atmospheric Effects

The atmospheric error (eR) is attributed entirely to the non-zero refractivity of the tro-

posphere. The troposphere is defined as the region of the atmosphere from sea-level to

about 70,000 ft. The refractivity (N) is a function of the atmospheric index of refraction

(n) defined in Eq (2-21). A non-zero refractivity (N) indicates an index of refraction (n)

greater than 1 and a decrease in the electromagnetic wave propagation velocity (v) from

the speed of light (c). This decrease in velocity results in an erroneous increase in the

line-of-sight range measurement. The basic equation is shown in Eq (2-22)

(14:6.3-1,6.3-3):

N = (l-n)x 106 (2-21)

1
v = - c (2-22)
n

The refractivity is a time varying quantity based on the atmospheric temperature (T),

water vapor content (e), and atmospheric pressure (P). This quantity is defined by Eq

(2-23) (14:6.3-4):

N- =77.6T(t)P(t) T(t) 2 e(t)

37,300 (2-23)

The refractivity at sea level, Ns, is assigned a nominal value of 330. In the CIRIS filter,

the refractivity above sea level is approximated with an exponential function based on

altitude above sea level (H), atmospheric scale factor (Hs), and sea level refractivity (Ns).

This equation is shown in Eq (2-24) (14:6.3-4).

2-12
 H

N = Ns eHS (2-24)

The CIRIS range correction (eR) is calculated based on a complicated function relating

refractivity, altitude, and length of measurement cycle. This algorithm is beyond the

scope of this description, but can be found in Reference (14:6.3-14,6.3-15).

This section outlines the correlated error compensation in the current CIRIS filter.

The open-loop corrections can be improved through the modeling of the uncompensated

correlated errors with stochastic shaping filters and allowing the Kalman filter to estimate

these errors. This idea is explored in later sections and implemented in the CIRIS "truth"

model developed in Chapter 4.

Kalman Filter Equations

The Kalman filter propagation and update equations are implemented in the MSOFE

simulation software described in the next section. The stochastic differential equation,

measurement equations, and residual equations must be defined to setup the Kalman filter

equations. In this section the form and notation used in the Kalman filter equations are

defined.

The stochastic differential equation must be defined first. The form of this equation

follows.

=
=(t) F(t) X(t) + W(t) (2-25)

E{ W(t)} = 0 (2-26)

E{W(t)W T(t+')1I = Q(t) 8,r) (2-27)

The states in a navigation stochastic differential equation are error states. Error states

(6X(t)) represent the difference between the true and the INS indicated states. Therefore,

the vector of states is represented by the 8X notation and the rate of change of the vector

of states by the X notation. The filter structure defined in this way is commonly referred

to as a linearized Kalman filter, where the F(t) matrix represents the linearization of the

error dynamics about the nominal state.

2-13
 The linearized Kalman filter estimate of the error states defined in Eq (2-25) is
defimed by a conditional mean, Eq (2-28), and a conditional covariance matrix, Eq (2-29).

E I6Z(t) =x(t) (2-28)

E{ [fX(t) - 8X(t)] [5X(t) - 8_(t)]lT IZ} = P(t) (2-29)

The expected value of 8X is the conditional mean of the error state (LX) and the conditionl

covariance of X is represented by P. P is a positive semidefimite n x n matrix (n=number

of states) that represents the statistical accuracy of the estimate. The square root of each

diagonal element of P is considered the standard deviation (1 a value) of the associated

error state estimate.

Mean = 51-K (2-30)

Standard Deviation = 1Iai = 4 (2-31)

i = 1,...,n

F(t) is an nxn matrix that represents the linearized model system dynamics. W is a

gaussian, "white" noise vector with a mean of 0, Eq (2-26), and a noise strength of Q,

Eq (2-27); Q is a positive semidefinite nxn matrix of noise strengths (16:275).

The measurement equation is defined next. The form of this equation follows.

8Z(ti) = Hk(t) 8X(t) + Vk(t) (2-32)

E{ Vk(t,)} = 0 (2-33)

E { Vk(ti) Vk(t))} = tj (2-34)

Since the stochastic differential equation is defined in error space, the measurement

equation is also defined in error space. 8Z represents the error in the measurement that

the Kalman filter is estimating. H is I xn measurement gradient matrix. V is a Gaussian,

"white" noise with a mean of 0, Eq (2-33), and a noise variance of R, Eq (2-34); R is a

positive definite scalar noise strength. The measurement equation is a scalar equation

and k ranges from 1 to the maximum number of different measurement types (16:275).

2-14
 The Kalman filter propagation equations are defined here. The continuous time
form of the equations follows (16:275).

8J(t/t,.j) = F(t) 8X@/rt. 1 ) (2-35)

P(t/ti.,) = F(t) P(t/ti.) + P(t/t,) FQr) + Q(t) (2-36)

The integration is started from the initial conditions:

-
P8(to) = Po
PO)= PO

After the first integration, the Kalman filter mean and covariance propagation equations
use the filter's last updated mean and covariance as initial conditions on the integration

performed to propagate the filter statistics to the next update point (16:275). These
propagation equations are integrated in MSOFE using a Kutta-Merson integration algo-

rithm (2).
The Kalman filter update equations are defined next. The standard form of these

equations follow (16:275).

K(ti) = P(t-) Hr(t) [H(ti) P(t-) HT (t) + R(t)] (2-37)

_(t,) = &X(t-) + K(t,) [8Zk(t) - H(t) 8_(tl] (2-38)

P(t'-) = P(t,-) - K(tt) H(t) P(t,-) (2- 39)

To increase numerical precision, these update equations are implemented in the U-D

covariance factorization form. To start the algorithm Po must be convened into its U-D

factors. The U-D factorization of P and the factorization equations are defined as (16:392):

P = U D UT (2-40)

For the n-th column:

D, = P,.,

{ = n-l,n -2.....

2-15
Then for the remaining columns, for j=n-ln-2,..., 1, compute:

= D U!

0 i>j

U 1 i=j (2-42)
[Pij- D U--+,
U UJ i=j-lj-2,...,l

The scalar measurement update for the U-D covariance factorized filter is defined as

(16:394):

f = UT (t,- ) Hr(ti)

vi = Djti-) f j 1,2,..., n

ao = R(t,) (2-43)

Then, for k=1,2,...,n, calculate:

ak = akI + fk Vk

Dkk(t,-) akI
ak

bk 4* Vk

fk
Pk ak-

Uj(t,)= Ujk(t-) + bj pk j =1,2,...,(k-1) (2-44)

b - bj + Ujk(t, ) vJ

(the symbol 4- denotes replacement by writing over) After the n iterations are com-

plete the updated mean and covariance are calculated with:

P(t,) = U(t,) DQ,) UT(ti+) (2-45)

K(t,) - (2-46)
a.

8X(ti') = WX(t) + K(t) [8Zk(rj) - H(t,) 8X(t-)] (2-47)

Finally, the residual equation is defined in the standard form as (16:394):

2-16
 Residual = Z(t) - H(ti) 8(tj-) (2-48)

Because the measurement is nonlinear, the extended Kalman filter form of the residual

is used to increase the accuracy of the residual calculation. This form of the residual

equation is defined as:

Residual = 64(ti) - 824,(ti) (2-49)

Here 8Z is formed based on the corrected INS and transponder data. The actual calcu-

lation of this residual is explained in Chapter 4.

Correlated Error and Residual Characteristics

The correlated measurement errors must be modeled with shaping filters and the

necessary states augmented to the dynamic system model. The shaping filters required

for this research are defined in this section (16:180-185).

The transponder survey errors result from limitations in the DMA geodetic survey

procedure. The resulting errors are constant in time and relatively small in magnitude.

In this research the survey errors are modeled for the X', Y', and Z' (E-frame) axes. The

calibration error in the aircraft transponder interrogator is also constant in time and rel-

atively small. These errors are "best" represented by a random bias (constant) shaping

filter. The defining stochastic differential equation is:

bE(t) = 0 (2-50)

6E(to) - O
Pa.t)= Po

The autocorrelation function and the power spectral density characteristics are shown in

Figure 2-3 (16:183).

A line-of-sight error occurs in the range measurement due to atmospheric conditions

changing the index of refraction and resulting in propagation delays. The resulting error

2-17
 AUTOCORRELATION POWER SPECTRAL DENSITY

(7) GCi
xx xx
[P +m
0 0

27Th[p +m 2
0 0

Figure 2-3. Random Bias Shaping Filter Characteristics (16:183)

is time varying and relatively small in magnitude. This error is "best" represented by a
first-order Markov process shaping filter. The defining stochastic differential equation

is:

6E(t) = _1 8E(t) + W(t) (2-51)

T

st(t 0 ) = mo

PSE(to) = PO

E{W(t)} = 0

E{W(t) W(+t+)} = 2 6() = Q 5(,)

T

This is an exponentially time-correlated process. The correlation time is represented by

the variable T and the initial variance by &" . The autocorrelation function and power

spectral density characteristics are shown in Figure 2-4.

2-18
 AUTOCORRELATION POWER SPECTRAL DENSITY

X - \ (61)
fX

+ 2 T2T

22
0.368 C 2

T lI/T

0y 2 el 2 C)2 T
(.2 2
+ (I/T)

Figure 2-4. First-Order Markov Shaping Filter Characteristics (16:183)

The Kalman filter residuals are a direct indication of how well the Kalman filter

model is representing the real world data. If the model perfectly matches the spectral

characteristics of the real world data, the residual data process will be a "white" Gaussian

sequence of mean zero and variance H()P(QHT(%) + R(t). Since it is a function of the

filter's covariance, it is independent of the real world measurements and can be calculated
"a priori". The square root of this covariance is designated the residual l value and is

referred to throughout the residual analysis section in Chapter 4 (16:229).

An example of a discrete-time "white" Gaussian noise with a 1y value of 1 is shown

in Figure 2-5. In the real world a perfectly "white" residual cannot be attained. However,

the closer the Kalman filter residuals are to this "white" process, with the applicable 1oa

magnitude, the better the Kalman filter model.

2-19
 3

-- I-

-2

-3

26000 25000 30000 32000 34000

SECONDS

Figure 2-5. Gaussian "White" Noise

Simulation Software
Several independently developed software simulation packages are used in this

research effort. All software packages are written in FORTRAN and used on a VAX

780, VAX 785, MicroVax I, or VAX 8650 Digital Equipment computer under the VMS

operating system. This section describes the major characteristics of the software package

and lists the respective references.

Multimode Simulation for Optimal Filter Evaluation (MSOFE). The

Avionics Laboratory at Wright-Patterson AFB, OH. and Integrity Systems at Winchester,

MA developed MSOFE (under contract to the Avionics Lab). It is a general purpose

multimode simulation program for designing and evaluating integiated systems based on

2-20
optimal (Kalman) filtering techniques. It provides two major operation modes: Monte

Carlo simulation and Covariance simulation. These two modes can be used independently

or simultaneously.

The user must set up 14 FORTRAN subroutines to use the package. Basically these
subroutines define the system ("truth") model, reduced order filter model, and the tra-

jectory for the time varying stochastic differential equations. These subroutines are
compiled and linked to the MSOFE main program to form the final simulation software.

An MSOFE control file is read upon execution to control and set up the specifics of the

user requirements for the simulation. The program outputs time histories of user selected
variables (means, covariances, measurements, etc.). The specific details on how to use

this package are beyond the scope of this introductory description, but can be found in
Reference (5:1-1,1-2,1-4).

The research conducted in this effort concentrates on the verification of the CIRIS
"truth" model. For this effort the MSOFE covariance and Monte Carlo simulation

capabilities are not required. However, the MSOFE software already contains all the
routines necessary to implement an extended Kalman filter with over 100 states; thus

the MSOFE package is chosen for this reason. The only major modification required is

to allow the filter to read measurements from an external data file; the filter model is

originally set up to receive its measurements from the internal system ("truth") model.

The only other supporting software developed is a program to convert the empirical

CIRIS flight tape into MSOFE trajectory and measurement data files. With these mod-

ifications, MSOFE is used to process empirical CIRIS flight tapes through a CIRIS full
ordered extended Kalman filter based on the CIRIS "truth" model. This same software

is also used to evaluate the reduced order CIRIS filter developed in this research.

Profile Generator (PROFGEN). The Avionics Laboratory at Wright-

Patterson AFB, OH developed PROFGEN. This acronym refers to a computer program

2-21
that is capable of computing the position, velocity, acceleration, and attitude rates for a

vehicle under continuous control moving over the earth's surface. This is done by solving

the kinematic equations of motion appropriate for a zero-mass body responding to

maneuver commands specified by the user. The user specifies a sequence of maneuvers

consisting of vertical turns, horizontal turns, sinusoidal heading changes, straight flights,

and rolls. The specific details on how to use this package are beyond the scope of this

introductory description, but can be found in Reference (1).

It is used in this research effort to generate the fighter flight profile required for the
verification of the INS "truth" model in Chapter 3. The MSOFE program is already set
up to read the flight trajectory files generated by PROFGEN.

MATRIXX. The term MATRIXx referes to a commercial analysis and simulation

package developed by Integrated Systems Inc., Palo Alto, Ca. This software package is

a powerful general purpose control design/synthesis simulation package. It features a

powerful matrix interpreter, a user-friendly environment with device independent

graphics, state-of-the-art numerical algorithms for reliable computations, and user-

transparent file management. The specific details on how to use this package are beyond

the scope of this introductory description, but can be found in Reference (22:30).

It is used in this research effort for supporting matrix operations. This consisted
mainly of data time history manipulations, filtering, and graphics. The Kalman filter

simulation software (MSOFE) was modified to output all data in a file format readable
by MATRIXx.

Summar.y
This chapter develops all the general theory applicable to the following chapters in

this thesis. The notation, coordinate frames, and coordinate frame transformation matrices

2-22
are developed in detail in the first section. The major operating characteristics of the

CIRIS Litton LN-39 INS and its Cubic transponder system are described in the second
and third sections. The Kalman filter notation and equations used are developed in detail
in the fourth section. The residual and correlated error characteristics are described in
section five. Finally, a brief introduction to the independently developed software sim-
ulation packages is given in section six.
The main purpose of this section is to present and explain the notation, equations,

and terms used throughout this research. No attempt is made to explain all the details
necessary for a complete understanding of the topic. However, appropriate references
are quoted for each major topic to allow the reader to further explore the material. The
additional detail required for a particular facet of this research is given in the applicable
chapter.

2-23
 III. INS "Tiuth" Model Design/Verification

This chapter describes the stochastic differential equations that define the LN-39
INS "truth" model. The equations presented are taken directly from the Litton systems

engineering report (21:6-9). Minor notational changes are made to increase the clarity

of certain variable names.

This "truth" model is coded into the filter differential equation routines of MSOFE.

The MSOFE simulations are conducted by propagating these differential equations over

different aircraft trajectories and plotting the time histories of the error state covariances.
The actual error state statistics plotted are the square roots of the diagonal elements of

the 85x85 error state covariance matrix. These values are referred to as the lF error state

covariance values throughout this section.

The results of a I0-run Monte Carlo analysis for a static flight profile and a simulated

fighter profile are documented in the Litton report (21:45-158). The MSOFE INS "truth"

model is verified by comparing the MSOFE 1a error state covariance value plots to the

Litton report's Monte Carlo l error state covariance value plots, for similar trajectories.

Error States
This LN-39 INS "truth" model consists of 85 error states. This error, 85 state vector

is partitioned into 6 sub-vectors which group similar error states. The total error state

vector is represented by SX. The subvectors are defined as follows: 8X, contains the 13

general INS dynamics errors such as position, attitude, velocity, and vertical channel

errors; 8X contains the 10 accelerometer, gyro, and barometer correlated errors; 5X

contains the 21 gyro bias errors; 6X 4 contains the 29 accelerometer bias and barometer

bias errors; 8& contains the 9 accelerometer and gyro initial thermal transient errors; and

6_X, contains the 3 gyro turnaround drift rate errors. The error states are all coordinatized

in the P-frame and the states in each sub-vector are shown in Table 3-1.

3-1
 The 85 error states and their initial l error state covariance values are defined in

Appendix B, Tables B-I thru B-9. The error states of main interest, that are continuously

referred to throughout this thesis, are defined here. They are the 80, SH, 64), and 8V

states. The angular error between the computer coordinate frame and the "true" coordinate

frame is represented by 8(. The angular error between the computer calculated altitude

and the "true" altitude is represented by 8H. The angular error between the platform

coordinate frame and the "true" coordinate frame is represented by 640. The error between

the computer calculated and the "true" earth referenced velocity defined in the P-Frame

is represented by SLV. The coordinate frames are defined in Chapter 2.

INS Stochastic Differential Equation

The INS time varying error dynamics are described by a single stochastic differential

equation.

8X(t) = FINs(t) 6X(t) + W(t) (3-1)

The partitioned 8X vector is defined in Table 3-1. In order to define the FINs matrix and

SW vector, they are also partitioned into smaller submatrices (Fi).

ax2
--
8 = 8X3 (3-2)
8X
8SX5

Fi, F12 F13 F14 F15 F, 6

0 F22 0 0 0 0

Fs= 0 00 00 00 00 00 (-3
(33)

0 0 0 0 F5 5 0
0 0 0 0 0 F6

3-2
 Table 3-1
Error State Partitions

Acc and Baro Gyro

General Correlated Gyro Bias Bias Thermal Turnaround
Errors Errors Errors Errors Transients Drift Errors
(13) (10) (9) (12) (12) (17) (9) (3)
Fxx

Fyy
~ABX z
66X
86 x ABY

Bxc Bx N2 ABZ FxzAXQ

O z By N AX AYQ

vx H SQAx
Bz xB z HYX FYP
- 1

8Gy X2 Hzy FQAZ BxB

8 HL 8Gz X3 Kxz SQX2 OY Bz

8Hc Kvz SQY, 0 2
8S3
84Kzx SQZ2 (Y3

8HB

SHsF

Note: See Appendix B-I thru B-9 for individual element definitions.

3-3
 - I

W- 0 (3-4)
0
0
.0

In order to solve this differential equation, the initial conditions on 8X and the initial

driving noise strength of W must be defined. The initial conditions required are the initial

error state means (X), state covariance matrix (P), and the initial noise strength matrix

(Q).
8X(t o ) =[] (3-5)

P1 , 0 0 0 0 0
0 P2 0 0 0 0
0 0 P33 0 0 0
- 0 0 0 P. 0 0
0 0 0 0 P55 0
0 0 0 0 0 P6

Q1, 0 0 0 0 0
0 Q22 0 0 0 0
= Qms(to)S@) 0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 00

The initial la error state covariance values for the states are defined in Appendix B,

Tables B-i thru B-9. The elements of the matrices F and Q consist of many different

physical variables which are defined in the next section.

Elements of the Fr Matrix

The variables shown in the elements of the INS dynamics matrix, FIN., (Appendix

B, Tables B-12 thru B-24) are defined in this section. The parameters A, f, and G. appear

3-4
in some of the following equations and they are defined in Chapter 2, Table 2-2. Elements

of the direction cosine matrix CE are also used and this matrix is defined in Chapter 2,

Eq (2-8).

The wander azimuth angle (x) is used in calculating the direction cosine elements

that are used in the definitions that follow. The best estimate of this angle is calculated

by subtracting the filter error estimate & from the INS a angle:

& = c S - (3-8)

where:

= -cos(ct) tan(L) 86 + sin(&) tan(L) 86, + 56)z (3-9)

The best estimate of the aircraft position vector components is defined as:

-(3- 10)

where:

[5Ll- sin(&) Cos(&) 6

0 [886[SX]
[j= cos(&)sec(L) -sin(&) sec(L) 0 O6y = 86 y (3-11)
S 00 1- Sri 1 M11
The best estimate of the aircraft velocity vector is formed by subtracting the filter estimated

velocity error from the INS velocity output, with components defined:

VINS, 8 X VX
[VNS Y -- 8y] = ry (3-12)
vINS,- Al- Jz

In the extended Kalman filter implementation, the best estimate of the position and

velocity trajectory variables are formed by subtracting the filter estimate of the posi-

tion/velocity errors from the INS position/velocity outputs. In the covariance analyses

that follow, only the error state covariances are calculated. Therefore, the trajectory

3-5
variables cannot be corrected by the filter error estimate. However, in these simulations,

the trajectory variables used are considered to be true values (without error) and do not

require correction.

The best estimate of the aircraft acceleration vector components are defined as:

A = JAIsj = AY (3-13)

Since the filter is not estimating the acceleration error, the best estimate of the aircraft's

acceleration vector is the INS acceleration output. The aircraft thrust acceleration value

is also required and is defined as:

A, = A, - G (3-14)

where:

G = Go 1-(2.00996)H + (5.28659E-3)(CE{3,3})2] (3-15)

1 ~A

The CE {i ,j} notation referes to the ij (row,column) element of the coordinate trans-

formation matrix. The components of the earth's spheroid inverse radii of curvature are

defined as:

I(
A[ l H
-A f[(C{3,3})2
f (FPC _2(CE{3, 1})2] ; P

1- -f[(C{33} -2({32})2] = [RY (3 6)

A(C {3, 1}) C{3,2} )

Three different angular rate vectors are required. The angular rate of the E-frame with

respect to the I-frame is defined as:

3-6
 Oi= Im (3-17)

= =[3,21%y (3-18)
.oCE{3,33}MZ

The angular rate of the P-frame with respect to the E-frame is defined as:

- -VCRZ -VYCRY P [%X -3_ 9

4E + YCZO~gP
=II (3-19)

The angular rate of the P-frame with respect to the I-frame is defined as:

ww
C
o)p+%C,2 j
1 1}
{3J (Lp (3-20)

The constants K, thru K4 represent feedback gains in the baro-altimeter vertical

channel aiding loops. They are defined as follows:

K, = (3-21)

2Go 4
K G. + 4 (3-22)

A2
K4 - z+A 2 (3-24)

where:

3-7
 A - HBARO

S= 100 1+

Ao,= 30 (INITIALLY)

(A.+8) IF ASA
A0 = (A-8) IF A 0 >A AND A.>38
30 OTHERWISE

The parameter P3 represents the inverse correlation time of the applicable state. They

are defined as follows:

1 SEC-' = 1 SEC-'
600 300

xccc -600 SEC- y 121520 SEC-

I SEC-' 3BxQ, YQ, Q, I SEC-'
3,,Q.,YQ~zo - 60 .60
1 SEC -'
PBXQZYQ2.ZQ2 140

where:

IVI X VY2+ VZ
V+

Elements of the 0 matrix

The variables shown in the elements of the Q matrix (Appendix B, Tables 10 thru

11) are defined in this section. The I3parameters shown in these elements are defined in

the previous section. The a values are defined as follows:

3-8
 looft aBxcc = 0.002 deg
CFSH =
hr

ZC = 0.005 deg gc.yczc = 3 g

hr-
deg
1, = 0.003 h- Y = 5 g

Alignment Simulation

The INS always undergoes an 8-10 minute gyrocompass alignment before being

placed in the navigation mode. This alignment process must be simulated before any

navigation simulations can be run. The Litton engineer responsible for the Litton INS

Monte Carlo simulations advised that the best way to simulate this alignment process is

to use a Kalman filter to update the INS with zero velocity measurements (2).

To accomplish this simulation, the measurement update routines in MSOFE are used.

The true trajectory variables L, X, H, Vx , Vy, Vz, Ax, Ay, and Az are constants and set

to 32.8 deg North, 106 deg West, 4200 ft, 0, 0, 0, 0, 0, and 32.08744 fps2 respectively.

The alignment simulation uses two velocity error measurements and the INS is updated

every two seconds for 10 minutes. The first scalar measurement is the Vx velocity error.

The H matrix (1 X 85) and value of the measurement noise R (1 X 1) are defined:

Hvx = [0 0 0 0 0 0 1 0 ... 0] (3-25)

ft2
Rv = 0.02 se (3-26)

The second scalar measurement is the Vy velocity error. The 1 X 85 H matrix and the

value of the measurement noise 1 X 1 R matrix are defined:

Hvy = [0 0 0 0 0 0 0 1 0 ... 0] (3-27)

Rvy = 0.02 -i (3-28)

sec-

3-9
 A covariance analysis is conducted using these measurements for a 10 minute period.

The resulting lo error state covariance values for the latitude, longitude, altitude, North

tilt, East tilt, azimuth tilt, North velocity, East velocity, and vertical velocity error states

are plotted in Figure 3-1.

In order to obtain the latitude and longitude errors in feet, the following transfor-

mation is used (3:48,94):

N = Tf pC TT (3-29)

where:

1,1} PINs{1,2}]
pC = P
[Prs{2, 1} P~s{2, 2)
= R sin(&) R cos(&x)]

I - R cos(L) R sin(a)J

R = A (1 - e sin2 (L))

The square root of the diagonal elements of the P" represent the 1c error state covariance

values of latitude and longitude in feet. To obtain the North/East tilts and velocities, the

following transformation is used:

[P]N = CN pP CP (3-30)

In this equation P represents the applicable 2X2 submatrix of PIs corresponding to the

X/Y tilts or velocities. The direction cosine matrix is defined in Eq (2-4).

The fully populated covariance matrix that results from this simulation is used as

the initial error covariance for all navigation simulations. The off-diagonal elements of

this matrix reflect the cross-correlation between the states as a result of the alignment.

The main purpose of this simulation is to obtain these off-diagonal elements. Producing

just the diagonal elements of the initial covariance matrix does not accurately portray the

resulting alignment correlations.

3-10
 4.5 460

4 400

1.5 360

300
FET3FE
25 FEETE25 ALTT D
2 200

15 150

100

5 50

0 100 200 .50 400 500 So0 0 0 100 200 300 400 500 600

SECONDS SECONDS

100 12000

10000

60
8=0

70 ARCSEC
ARCSEC
60 VERTICAL TILT
4=0

40 2000

NORTH TILT
30 I. 1 1. . 0
0 100 2oo 500 400 300 6w 0 100 200 00 400 5w, 600

SECONDS SECONDS

.12 4

3.5

.13

06
2.5
FPS
.6NORTH VELOCITY FpS 2

.04 ~ IsVERTICAL VELOCITY

02
EAST VELOCITY

0 o w 00 100 200 300 400 500 60

SECONDS SECONDS

Figure 3-1. RMS Errors for Alignment Simulation

3-11
Aircraft Trajectory Simulations
A covariance analysis is conducted with the INS model using two different aircraft

trajectories. In these simulations the initial covariance matrix generated by the alignment

simulation is used. This simulation propagates the covariance of the error states without

any measurement updating (deterministic baro-altimeter aiding is included in the INS

model). The results of this simulation define the baseline error growth for the unaided

INS subjected to the two respective aircraft trajectories.

Static Profile. In this simulation, the true trajectory variables L, X, H, V x , Vy,

V z , A x , Ay, and A z are constants and set to 32.8 deg North, 106 deg West, 4200 ft, 0, 0,

0, 0, 0, and 32.08744 fps 2 , respectively. MSOFE is used to propagate the stochastic

differential equation over a 10 hour period. The lao error state covariance values of

interest for this simulation are shown in Figure 3-2.

The latitude, longitude, North tilt, East tilt, North velocity, and East velocity lo

error state covariance value plots are compared to the corresponding Litton Monte Carlo

I error state covariance value plots in the Litton report (21:79-82). The shape and

magnitude are very similar, indicating the MSOFE INS error dynamics are properly

modeled.

Fighter Profile. In this simulation, the true trajectory variables are generated

using the PROFGEN trajectory simulation software. The fighter profile defined in the

Litton report (21:41) is approximated as closely as possible. Due to the complexity of

this profile, it could not be exactly replicated. However, it is approximated sufficiently

for the INS error growth to be similar. This 2-hour fighter profile is shown in Figure

3-3.

3-12
 4.5 278

* 2786

.SLA TITUDE 278 ALTITUDE

277.5

NA(7TCAL 25D ET277

276.5
2
276

18 .- \275.5
LONGITUDE27

275,

................ 27.5

8 I ~ 1 1 .24I 1

HOURS HOURS

42 220

41.5 210

40 t
190S VERTICAL TILT

ARCSECS Igo VM IT
0
NORTH TILT EATTL
40s

160

ISO

W5 40

HOURS HOURS

3 .525

2.7
.592
VERTICAL VELOCITY
2.4
.5915
21

591
18
FS NORTH VELOCITY FPS

1.2
.59

3 EAS VELCITY589

0 .0..
. . l..
!,. I ... .... 588 r

HOURS HOURS

Figure 3-2. 1 a Values for the Static Simulation

3-13
The simulation software (MSOFE) is used to propagate the stochastic differential equation

over the 2-hour trajectory period. The resulting l error state covariance value plots are

shown in Figure 3-4.

The latitude, longitude, North tilt, East tilt, North velocity, and East velocity la

error state covariance value plots are compared to the corresponding Litton Monte Carlo

la error state covariance value plots in the Litton report (21:105-108). The shapes and

magnitudes are very similar, thus validating the MSOFE INS error model.
Summary

The stochastic differential equation and all associated parameters are defined in

detail. The 10 minute gyrocompass alignment is simulated to generate an accurate initial

covariance matrix for the navigation simulations. Finally, a MSOFE covariance simu-

lation is conducted for the static and fighter profiles defined in the Litton report. The

results of the MSOFE simulations closely agree with the Monte Carlo simulations in the

Litton report. Therefore, the INS model is considered to be properly modeled and coded

into MSOFE.

3-14
 - I

-- --- - -- IALTITUDE
-

(FEET)

S~ 0000.

4~0000.

20000.

10000.

0.

LONGITUDE 34 LA TITUDE
(DEG)-105.(DEG)
32.

Figure 3-3. PROFGEN Fighter Profile

3-15
 * 2000

LTWDEE 10

7 1900 ALTITUE

*~ 400

NALITICAL FEET
1.200
mu

4 1000

3LONGITUDE40

0200 L ...
'I 20 40 s0 so too 120 0 20 40 s0 so 100 12C

MINUTES MINUTES

40'a 158

40.5 - EAST TILT 14

40.2 - 152

329 - 150

321 148

ARCSECS 393-ARCSECS 16 VERTICAL TILT

39 144

NORTH TILT
3417 142

394; 140

38.1 -. 136

0 20 40 60 80 3
2. 0 20 40 80 so too 20

MINUTES MINUTES

24

2.1 NORTH VELOCITY

18a6

VERTICAL VELOCITY

FPS FPS
1.2

EAST VELOCITY
I2

0 . L 0 100.. 120.
0 20 40 60 so 0 o0 2 00 4 120 0 3

MINUTES MINUTES

Figure 3-4. 1 cy Values for the Fighter Simulation

3-16
 IV. INS/Transponder "Truth" Model Design/Verification
This chapter defines the range and range-rate measurement models. It augments

these models to the INS "truth" model to form the full-ordered extended Kalman filter

model. The full-ordered filter is used to process empirical data from a CIRIS flight. The

results of this simulation are analyzed in detail and compared to the data from the current

CIRIS filter.

Range Measurement Model

Since this is a linearized Kalman filter, the measurement input to the Kalman filter

is the error in the range measurement. This quantity is formed by differencing the cal-

culated range and the range measurement available from the Cubic RRS.

The calculated range is based on the aircraft's INS position data and the transponder

position data. This measurement is depicted in Figure 4-1. The INS and each ground

transponder position vector are defined in the ECEF frame as follows:

PA = YA
_ZA

The A and T subscripts indicate aircraft position and transponder position respectively.

The nonlinear range measurement based on these positions is defined as:

RJEA,PT) = IPT- I = (XT-X1) 2

+ (YT--y) + (Zr-ZA2 (4-1)

The equation defining the error in this measurement is obtained by performing a Taylor

series expansion of Eq (4-1) about the true values. This results in:

4-1
 - - + XcTA I T)

a1%_P_$.n£e)
+ IPT) 1.4Tr-) PT + H.O.T. (4-2)

Truncating to first order and expanding the partials results in:

XT - XA _YT - Y- Zr - Z
Rc RRc6X
=c R ^

X - XA XT+ YT - Y 7-, - ZA 8Z (43)

Z' E-FRAME

INS
E E, AIRCRAFT
C 1-
cA POSIMON
R VA

TRANS PONDER TRUE

POSMONAIRCRAFT
s6ZPOST ON E) X

POSMO T / 'qq
TRANSPONDER - ---------------- ---------
,,. - -.... __-
---- ---- --- ---- ---

Xf//

Figure 4-1. Errors in Calculated Measurement

The Cubic RRS measured range contains the true range plus an error due to a RRS

calibration bias (6EcA), an error due to atmospheric propagation delays (&EArT), and a

random error (VR). This measurement is defined as:

4-2
 RM = R uy + + EATM - VR (4-4)

The measurement to the Kalman filter is formed by differencing Rc and RM. This

differencing cancels the true range quantity and leaves only the error which the Kalman
filter is estimating. This measurement is defined as:

Z = 8ZR.NGE = Rc- RM =

XT - XA YT -- Y Z ZA
R XA R R SZA

XT - XA _ -_+_YT YA+ Zr - ZA

R T R T R
- - 8EATM + VR (4-5)

The stochastic differential equations are now developed for the errors shown in Eq

(4-5) that are not already modeled in the INS "truth" model. The INS (vehicle) position
6
errors (SXA ,AY, ZA) are the only errors already modeled. The transponder position
errors (5XT, 'Y, M-r) are modeled as random biases with an initial mean of 0 ft and
initial 1y value of 5 ft. The describing stochastic differential equation is:

kP_(t) = 0 (4-6)
a,_to) = o0
Par(to) 0 25
02 0 2

The Cubic RRS calibration error is modeled as a random bias with an initial mean of 0

ft and initial lo of 5 ft. The describing stochastic differential equation is:

4-3
 AEPcA(t) = 0 (4-7)

'5fCA.(to) = 0

PW.C ,(to) = I

The Cubic RRS atmospheric error is modeled as a first order Markov time varying error
with a correlation time of 300 seconds, noise strength of 0.1667 ft2/sec, initial mean of 0

ft, and initial la value of 5 ft. The correlation time for this model is derived from static

testing data recorded at CIGTF. Range data was recorded between transponders separated

by different distances. After removing the true range, the power spectral density (PSD)

on this error data indicated an average correlation time of 300 seconds.

'EATt(t) =- I SEAT(t) + WATm (4-8)

300
8,LQt o) = 0

Pw,,(to) = 25

E{WATt) = 0

E{WATt)WAT(r+T)} = 0.1667 8(T)

The random measurement noise (VR) is modeled as Gaussian, "white" noise with a mean

of 0 ft and a noise strength of 4 (ft) 2 (RR). These models are augmented to the INS "truth"

model to form the final model for the full-ordered Kalman filter. The measurement

gradient matrix (HR) is taken directly form Eq (4-5). The details of this process are shown

in the "Full-Ordered Kalman Filter "Truth" Model" section that follows.

Range-Rate Measurement Model

Again, the measurement input to the linearized Kalman filter must be the error in

the range-rate measurement. This quantity is formed by differencing the calculated

range-rate and the range-rate measurement available from the Cubic RRS.

4-4
 The calculated range-rate measurement is based on the INS position and velocity

data. The nonlinear range-rate measurement equation is obtained by differentiating Eq

(4-1). Ibis results in:

dRC -_(XT-XA) VAX - (YT-YA) VAY - (ZT-ZA) VAZ (49)

R dt Rc

The nonlinear equation defining the error in this measurement is obtained by performing

a Taylor series expansion on Eq (4-7). This results in:

=c P" + aPA LR,.VA(TnRPe _

+ DOCA IA._
YArh) 8VA + H.O.T. (4-10)

Truncating to first order and expanding the partials results in:

RC = ftRU
2
(XT-XA) VAx + (XT-XA)(YT-YA)VAY + (XT-XA)(Z'-ZA)VAZ
............
. - VAXR
.. 8X^

2 -(YT- YA) (r-ZA)VAZ- VAYPK

(YT- YA) VAY + YA) (XT-XA)VAx + (YT- BY^

(r-ZA)VAz + (Zr-ZA)(XT-XA)VAx + (Zr-ZA)(YT-YA)VAy - VAzR

- 8ZA

(XT
-8Vx- - XA) (YT - YA) 6V^AYz(4-
(ZT- ZA) AZ (4-11)
1
Rc Rc Rc

The Cubic RRS measured range-rate contains the true range-rate, an error due to a

RRS calibration bias (EcAL), and a random error (VRR). This measurement is defined as:

RM = RUE + 8EcA" - VRR (4-12)

The measurement to the Kalman filter is formed by differencing Rc and RM. This

differencing cancels the true range-rate quantity and leaves only the error which the

Kalman filter estimates. This measurement is defined as:

4-5
8Z 2 = = - RM =

(XT-X)VX + (XT-XA)(YT-YA)VAY + (XT-XA)(ZI- ZA)VAz - VAXRk

(YT- YA)2VAY + (YT-YA)(XT-XA)VX + (YT-YA)(Zr-ZA)V A - VAXl

(Z- ZA) 2VAz + (ZT-Z)(XT-X)VAX + (ZT-ZA)(YT-Y)VAY - VAZRK

(XT - XA) (YT - YA) - (Zr- ZA)

-
Rc SVAx - _Vy
Rc __A
Rc

- 8EcAL + VR (4-13)

The Cubic RRS calibration error is modeled as a random bias with an initial mean of 0

fps and initial uncertainty la of .1 fps. The describing stochastic differential equation

is:

8Ecaa (t) = 0 (4-14)

tCAUM(to) = 0

P MC (to) = 0.01

The random noise (V.) is modeled as Gaussian, "white" noise with a mean of 0 fps and

a noise strength of 0.09 (fps) 2 (R.). These models are augmented to the INS "truth"

model to form the final model for the full-ordered Kalman filter. The measurement

gradient matrix (H.) is taken directly form Eq (4-13). The details of this process are

shown in the "Full-Ordered Kalman Filter "Truth" Model" section that follows.

Full-Ordered Kalman Filter "Truth" Model

The additional states required to model the correlated errors in the range and

range-rate measurements are augmented to the INS "truth" model stochastic differential

equation to form the full-ordered "truth" model stochastic differential equations. The

calibration range and range-rate random biases are assumed to be in the interrogator on

4-6
the aircraft. Therefore, these errors are common to all the transponders interrogated and

require only 2 states. The transponder survey errors and atmospheric error are unique to

each transponder. Therefore, each transponder interrogated requires 4 unique states.

A typical CIRIS flight uses from 10 to 30 different transponders. It normally

sequences through a 5-10 transponder window at a 1 Hz rate. It deletes/adds transponders

to this window to optimize the transponder geometry. Providing 4 states for each

transponder is computationaly burdensome. To keep the computational load manageable,

4 states are provided for only 10 different transponders (maximum number possible in a

window) for a total of 40 additional states. When a transponder is deleted and replaced

with a new one, the four rows and columns of the covariance matrix for the respective

transponder error states are set to zero. The mean for each of the respective states is set

to zero when the transponder is first used in the flight. However, the algorithm saves the

mean of the current transponder before switching to a new transponder and if this

transponder is encountered again in the flight, it reinitializes the error state means with

the saved values. The covariance diagonal element for each of the four states is rein-

itialized to its initial value of 25 ft2 . With this procedure, any number of transponders

encountered during a flight are accommodated without additional states.

The 127 (85+42) state augmented stochastic differential equation is:

XQ) = F(k-,t) X(t) + W(t) (4-15)

8X-LXNs
= j
F =
F=[As'j 0L
s 0-W
--
WIS=
]
S.'0 0) = [9 P8,~to) = PINS 0

E= 0

0
E{W(to) WT(t + C)} = Q(to) S(T)= QS 8(0

4-7
The variables designated INS are defined in Chapter 3. Note that the F matrix is calculated

based on X which is the best estimate of the true states. This full state vector is def'ned

as:

= X - '59 (4-16)

The variables subscripted "MEAS" are defined in Appendix C.

The measurement equations are defined as:

8Z,(t,) = 8ZlANGE(ti) = HR('P,t,)S X(t,) + VR(t,) (4-17)

-A LA

-TPT - P
E{VR(t,)} = 0

= ti tj
E{ VR(ti) VR(tj) }

Z 2(t,) = SZ =(t) H TR1Y__

,X,(t,) + VRR(t,) (4-18)

VA =V -8A

E{ VR(t,) = 0

E {Vmx(t,) VRR(t ) } -- t t
0 t' #tj

Eqs (4-17) and (4-18) are scalar equations. Again note that the best estimate of the full

states are used to calculate the measurement gradient matrices (HR and H.). The HR

and HR matrices are defined in Appendix C. The Kalman filter is updated with these

equations, for a single transponder, every second.

The last quantity required to execute the Kalman filter algorithm is the residual.

Because the measurements are nonlinear, the extended Kalman filter form of the residual

equation is used. This form is defined as:

4-8
 RES(t 1 ) = c(P_A, T t,) + 81cAta(t) + 8tATM(t) - Rm(ti) (4-19)

RES2 (r,) = (PAT' A'tt) + 8tcA1(t) - R1Mt) (4-20)

The range (Eq (4-1)) and range-rate (Eq (4-9)) nonlinear equations are used with the best

estimate of the full states. This form of the residual calculation takes into account the

nonlinearities of the quantity and results in a better estimate of the true residual. This is

in comparison with the linearized error residual update form (15)

CIRIS Flight Simulation

A CIRIS flight tape was provided by CIGTF. This data tape contains the INS

position, INS velocity, baro altimeter, range measurement, and delta range measurement

data for the flight. The tape also contains the error state means, error state covariances,

and measurement residuals from the current CIRIS filter for this flight. The flight tra-

jectory for this data is illustrated in Figure 4-2.

The flight is approximately 4.5 hours in length. CIRIS flight time is designated in

seconds; it begins at time 26132 and ends at time 42000. Since the full-ordered filter is

computationally intensive and because most of the high dynamic maneuvers occur early

in the flight, only the first 2.25 hours of the flight are processed by the filter; beginning

at 26132 and ending at 34000. The T=34000 point is shown in Figure 4-2. This simulation

required 17 hours of CPU time on a VAX 8650 computer. The latitude vs longitude and

altitude vs longitude position trajectory plots, including transponder locations, for the

abbreviated CIRIS flight profile are shown in Figure 4-3.

The data from 3 transponders is selected for a detailed residual and correlated error

analysis. These transponders are designated 181, 185, and 186; there locations are circled

in Figure 4-3.

The INS acceleration data is not available on the CIRIS flight tape, but is required

for the full-ordered Kalman filter INS error dynamics model. This data is obtained from

the INS velocities using the following equations (21:34,37):

4-9
 25000.

20000.

ALTIUME
15000.
(FEE)

10000.

5000.

-114.

32.0
32 OIGIN LONGnIUDE
T2. 2613 2 -- 08.(DGES
LAI"][UDE 33 . 0(DGE)

(DEGREES) 33..5 -105.

Figure 4-2. CIRIS Flight Trajectory

AVx
At + - + )+ (4-21)

AVv
z o') V x - (o), + %r) Vz (4-22)
At + (o +

Az - At J (ow + woi) V - (o., + wr ) Vx + G (4-23)

The full-ordered filter also requires range-rate measurement instead of the delta range

measurements provides on the CIRIS flight tape. The delta range measurements are

converted to range-rate measurements using the following equation:

RANGE RATE = DELTA RANGE (4-24)

TM - TR

TR = Start of Measurement Cycle

TM = End of Measurement Cycle

4-10
 1191
3.. x 187
Ix 202

33 2

.33
1988
193 8 197
xx
32.6 18

32.4 -193
32-T.34*m
32 2DEGRADED TRANSPONDER I EXCWLLANT TRANSPONDER
COVERAGE ICOVERAGE
32 . I . . I . - - - I .1I ,
-108 5 - 108 -107.5 -107 -106 5 -106 -05

27000

2.4000

21 000

1 8000

198 185019

62000

9000 x

3000 iS OR0G2

173 T=241SZ

-00 -7.!5
11D. -107 -10e. 5 -105 -105.5

LONGIT'UDE (D)EG)

Figure 4-3. CIRIS Abbreviated Flight Profile

The measurement cycle times (T. and TM) are available on the CIRIS flight tape.

A FORTRAN program is used to preprocess the CIRIS flight tape. This program

generates the acceleration data using Ecvs (4-21), (4-22), and (4-23). It converts the delta

range measurements into range-rate measurements (Eq(4-24)). Finally, the programs

writes the flight trajectory data (INS position, velocity, acceleration, and baro altimeter)

4-11
to an MSOFE flight file, and the range and range-rate data to a MSOFE measurement

data file. This INS data and the transponder mapping for this flight are shown in Appendix

D.

These data files are processed in MSOFE from takeoff at time 26132 to time 34000

(2.25 hours). The range and range-rate measurements are updated from 1 transponder

every second. A 70 tolerance is used for residual monitoring. Any measurement greater

than 7 times the predicted residual a is rejected for both range and range-rate measure-

ments. This residual monitoring technique resulted in 14% of the 16,060, range and

range-rate measuremcnts from all transponders, being rejected by the filter; this number

of rejected measurements is considered acceptable and does not affect the overall quality

of the filter estimated errors. The error state means, error state covariances, and residuals

are analyzed in the following sections.

Residual Analysis

The range and range-rate residuals are a direct indication of how well the Kalman

filter model is representing the real world. If the Kalman filter model matches the real

world model, the residuals become a "white" noise process with a mean of zero and a

1 a value equal to (HPHT + R)2 . In most cases it is not possible to model the real world

error sources, therefore the correctness of the model is directly related the residual sta-

tistics proximity to the ideal statistics.

The current CIRIS residuals are also examined to compare and emphasize any

differences. The range and range-rate residuals are analyzed for transponders 181, 185,

and 186. The locations of these transponders, with respect to the flight trajectory, are

shown in Figure 4-3 (locations are circled). The current CIRIS filter uses delta range

measurements instead of range-rate measurements, but since the two quantities are linearly

related (Eq (4-24)), the overall envelopes should be similar. The full-ordered filter

residuals are designated "TRUTH" and the current CIRIS filter residuals are designated
"CIRIS."

4-12
 The range residuals are shown in Figure 4-4. The left column contains the range

residuals for the full-ordered filter and the right column contains the range residuals for

the current CIRIS filter. The 1 a values shown in the plots are the filter calculated

(HPH " + R) values.

The predicted and actual means and 1 a values for the full-ordered filter and the

current CIRIS filter range residuals are shown in Table 4-1. The full-ordered filter

residuals fit the predicted 1 a very well, but the current CIRIS filter residuals do not fit

the predicted 1 a values. The current CIRIS filter residuals are also biased and exhibit

correlated errors. This indicates the full order filter is doing a much better job in estimating

the correlated errors.

Table 4-1
Range Residual Statistical Comparison

"Truth" CIRIS

Transponder Mean la Value Mean la Value

ft ft ft ft

181 +0.14 6.56 +1.97 6.49

185 +0.28 4.58 +11.02 5.51

186 -0.04 6.09 +6.55 6.74

Filter Predicted 0.00 3.0 0.00 16.0

The actual I a values for the "truth" and CIRIS filters are very close. This indicates

that the magnitude of the random error in both filters is the same, as expected. Overall,

the "truth" model filter range residual characteristics are closer to the ideal (filter predicted)

residual characteristics. Therefore, the position and velocity estimated errors predicted

by the filter based on the "truth" model are closer to the real world true errors than the

current CIRIS filter error estimates.

4-13
 30 30

20 20 RANGE RL5IDAL (CIRIS)

10 10

FEET 0 FEET 0

-10 -10

-20 -- 20
T.I8I c
RANGE RESIDUJAL ("TRUTH")
30 . I . . I . I . IJ-30

SECONDS SECONDS

30 30

20 -20

10 10

FEE 0 FEET 0

-0 -- 10

-20 2 1c

RANGE RESMDAL ("TRUTH") RANGE RESIDUAL (CIRIS)

27000 21000 2900 30000 3 1000 300 40
Mom 27D00 21000 29000 30000 3100D 32000 33000 34000

SECONDS SECONDS

30- 30

2001c/ 2

10 - o0

FEET 0 FEET 0

-'0 -10

-20 -- 20

-3 ANGE REIDVAL ("TR~flI") 210 t40

0 000 200 340
210 00 300 32000 3o S0 110 00 2M M0

SECONDS SECONDS

Figure 4-4. Range Residuals (CIRIS and "TRUTH")

4-14
 The range-rate residuals are shown in Figure 4-5. The left column contains the

range-rate residuals for the full-ordered filter and the right column contains the delta range

residuals for the current CIRIS filter. The 1 a values shown in the plots are the filter
calculated (HPHT + R)' values.

The predicted and actual means and 1 a values for the full-ordered filter and the

current CIRIS filter range-rate residuals are shown in Table 4-2. The full-ordered filter

residuals fit the predicted I a very well, but the current CIRIS filter residuals do not fit
the predicted 1 a values very well. The current CIRIS filter residuals are also biased and

exhibit correlated errors. Again, this indicates the full order filter is doing a much better

job in estimating the correlated errors.

Table 4-2
Range-Rate Residual Statistical Comparison

"Truth" CIRIS
Transponder Mean la Value Mean 1a Value
fps fps ft ft
181 +0.07 0.48 -0.08 0.20
185 +0.05 0.43 -0.14 0.18
186 +0.01 0.47 -0.12 0.20
Filter Predicted 0.00 0.3 0.00 0.3

The actual I a values for the "truth" and CIRIS filters are significantly different in

this case. This is due to the residual calculations being based on different equations and

presented in different units (ft vs fps). Basically, the delta range measurement is the

integral of the range-rate measurement over the applicable measurment cycle. Therefore,

this difference is reasonable. Overall, the "truth" model filter range-rate residual char-

acteristics are closer to the ideal (predicted) residual characteristics. Therefore, the

position and velocity estimated errors predicted by the filter based on the "truth" model

are closer to the real world true errors than the current CIRIS filter error estimates.

4-15
 252.

2 2

I .5 '.5

FFS50 FEET c)

t Cr -1.5

T-181
RAN AT 18 -2D± DELTA RANGE RESIDUAL (CIRIS)
-25 RNFR7R A 7Ur ...- 25: I-
26000 25000 20000 32000 24000 26000 28000 20000 52000 34000

SECONDS SECONDS

2 2

5I c

"S 0FEET 0

- -- -- - -

2 -T-185 -2 T-185
RAG
AERESDAL e7TRU ) I DELTA RANGE RESIDAL (CIRIS)
25 -25 ... _________I__...._________
2 70D0 28000 29000 30000 31000 32000 33000 34427000 28000 26000 30000 31000 32000 32000 34000

SECONDS SECONDS

25 2.5

2 2

1.5 10
1. 1

"S 0L V FEET 0

-2.5 -25.1. I . . I .1
-lo0om00 32000 34lu0 2400 2600 30000 IV00 34=0
SECONDS SECONDS

Figure 4-5. Range-Rate Residuals (CIRIS and "TRUTH")

4-16
 Overall, the "truth" model residual statistics are closer to the filter predicted than

the current CIRIS filter. The measurement white noise strengths (RR and RRR) could be

adjusted to draw the "truth" model filter residual statistics even closer to the predicted.

However, this process is very time consuming and the possible gain in accuracy is not

significant enough to warrant this additional time. This topic is discussed further in

Chapter 6.

Full-Ordered Filter Data

The full-ordered filter's means and I Yvalues for the position and velocity error

states are examined first. Next, the differences between the full-ordered filter and the

current CIRIS filter means and 1 ; values are plotted to compare the two filters.

Means and 1 a Values. The position error state means and 1 a values are shown

in Figure 4-6. The INS latitude error growth increases from 0 to approximately -2700 ft

in the 2.25 hour period and the filter's assigned 1 a value varies from 2.5 to 8 ft; note

that the Schuler mode is clearly visible in the error growth. The longitude error growth

increases from 0 to -3000 ft and the filter's assigned I a value varies from 3 to 6.7 ft.

The altitude error growth (ignoring initial transient) varies from 0 to -900 ft and the filter's

assigned 1 a value varies from 10 to 90 ft. The general shape, magnitude, and sign of

the "truth" model filter's position error means follow closely the current CIRIS filter

means. The increasing position error is characteristic of the INS. The 1 a values of the

"truth" model filter position error estimates are less than the current CIRIS position error

I a values; indicating the higher accuracy in the "truth" model position error estimates.

The areas of lower accuracy in the I a plots corresponds with the degraded transponder

coverage areas shown in Figure 4-3.

4-17
 0

-300 LTITUDE ERROR

-600 /
-1200 FEET6
FEET

-1800

-21D00 4

-2400 LAT1TUDE E ROR

(MAN) 3
-27MO

-m 80 2000 00 540W 260 280 300 20 40

SECONDS SECONDS

06

-500

0 LONGrTUDE ERROR
-00LONGITUDE ERROR (I0')
(MEAN) 5

FEET FEET
-2000 As

-2500
3.5

-2000

-3500 I 2,5

SECONDS SECONDS

1500 - o0
ALTITUDE ERROR
ALlTTUDE ERROR
1200 MAN) OD (1'
gm00 70J

600 60

30D FEET 0

FEET 0 40

-300 - 0

-600 20

-00 I0

-1200 0
200 2S000 20000 32000 3400 6000 2 600
28 00M0 320D0 34000
SEtCONDS SECONDS

Figure 4-6. Position Error States ("TRUTH")

4-18
 The velocity error state means and I a values are shown in Figure 4-7. The North

velocity error growth varies from +0.6 to -1.2 fps and the filter's assigned I a value varies

from 0.025 to 0.057 fps. The West velocity error growth varies from +0.6 to -1.2 fps

and the filter's assigned 1 a value varies from 0.035 to 0.057 fps. The vertical velocity

error growth varies from +3 to -3 fps and the filter's assigned I a value varies from 0.1

to 0.51 fps. The general shape, magnitude, and sign of the "truth" model filter's velocity

error means follow closely the current CIRIS filter means. The I a values of the "truth"

model filter position error estimates are less than the current CIRIS velocity error 1 a

values; indicating the higher accuracy in the "truth" model velocity error estimates.

Means and 1 a Value Differences. The difference in position and velocity

error state means and I ar values between the full-ordered filter and the current CIRIS

filter are shown in Figure 4-8 and 4-9 respectively. The differences are formed by

subtracting current CIRIS filter data from the full-ordered filter data ("TRUTH" - CIRIS).

A large initial transient is exhibited in all the difference plots. This large initial error is

caused by the CIRIS flight tape recording start time. Due to delays before takeoff of the

aircraft, the flight tape recorder is not started until just before takeoff. A significant

amount of time can elapse between the end of INS alignment and the tape turn on. The

INS data at the beginning of the tape can already contain substantial error growth. The

initial transient is the amount of time required by the full-ordered filter to estimate this

error. The current CIRIS filter does not experience this anomaly because it is operating

in real time and receives all the data starting from the end of INS alignment. Therefore

this initial transient is ignored in the analysis that follows.

4-19
 9 07

NORTH V9LOCITV RROR NORMh VELOCT ERROR

ME4AN) O'

3
055

FPS ~FPS 0
-3 045

04

035

-02

t .02

260 28000 000 32000 34=0 25000 2B000 .50DOO 32000 34000
SECONDS SECONDS

6 06

3 055

o 05

-304

FPS
Fps 04

035

2 03

-15 WEST VELOCITY ERROR 025 WEST VELOCITY ERROR

(MEAN)(0'
-18
5 . I 02 . . .
25000 MOO. 30000 32000 300 25000 00 W0ON) 320D0 34000
SECONDS SECONDS

4 6

VERTICAL VELOCITY ERROR

VERTICAL VELOCITY ERROR (1I'
3-MAN) 5

UPS FF5 .

-22

SUCO Zw00 30000 32000 54000 028000 2500 3000 300

SECONDS SECONDS

Figure 4-7. Velocity Error States ("TRUTH")

4-20
 The position difference tine histories are shown in Figure 4-8. These plots reflect

the time varying characteristics of the filter position error state estimate difference. This

magnitude of the difference is highly dependent on the aircraft trajectory and transponder

geometry; in addition to the differences in the filter modeling of the correlated errors. To

quantify the differences the mean and standard deviation is calculated for the position

means and la values. These position difference statistics are shown in Table 4-3. The

magnitude of the average difference is less than the independently verified position

accuracies of the current CIRIS filter. This average mean differer.ce is representative of

how much the "truth" model filter's position error state means are moving towards the

true error.

The velocity difference time histories are shown in Figure 4-9. These plots reflect

the time varying characteristics of the filter velocity error state estimate difference. This

magnitude of the difference is highly dependent on the aircraft trajectory and transponder

geometry; in addition to the differences in the filter modeling of the correlated errors.

Table 4-3
Position Error State Differences

Mean Difference l Value Difference

Position Mean 1 Value Mean 1y Value
Error States ft ft ft ft
Latitude +3.37 7.98 +0.03 0.42
Longitude +5.73 10.29 0.00 0.39
Altitude +25.66 63.86 +6.82 5.04
Note: All differences are "Truth" - CIRIS

To quantify the differences the mean and standard deviation are calculated for the

velocity means and la value differences. These velocity difference statistics are sho - I

in Table 4-4. The magnitude of the average difference is less than the independently

verified velocity accuracies of the current CIRIS filter. This average mean difference is

representative of how much the "truth" model filter's vei .-city error state means are moving

towards the true error.

4-21
 404

30 I LA7ITUDE ERROR DIFFRECE

1'

207

2
'0

FEE FFEET0 i fI.IT0FE

I

-20 0 N
-30

-20 (MEAN)

-40 L.ID ERO

. D.FFERE-2E
26000 ~ 00 1000 32000 _"ODD 26000 26000 032300
SECONDS SECONDS

210 3

'60 LOGTD RORDFEEC

LONGTUD
DFFERNCELONGITUDE
ERRR ERROR DIFFRENCE
150 (MEAN) 2.5

'20 2

FEET 60 15
FEET
601

30

0
0

-30 -. 5

SECONDS SECONDS

210a 60

1600 50
ALTITUDE ERROR DIFFRENCE ALTITUDE ERROR DIFFERENCE
OMAN)(1)

1200 30

FEET FEET
too 20

600 10

500 0-7

6 o -10

-300I -20 I

SECONDS SECONDS

Figure 4-8. Position Differences ("TRUTH" - CIRIS)

4-22
 a 01
NORTH VLOCITY ERROR DIFFERENCE
0

-. 01

4
-.02

2 -03

0 -. 04

IV0I
1171
-2

-. 07
NORTH VELOCITY ERO DIFRNCE

26000 2800 300 2000 34000 26000 28000 30000 32000 34=0
SECONDS SECONDS

-0i
WEST VELOCITY ERROR DIFFRENCE
.3 --. 02 I '

.2 -. 03

0
FPS - 06
FS-.06 -I
-. 1 _ 0N

-2 -. 07

-. 3 -. 06

-4 WEST VELOCITY RR RENCE -.09

2600 2&00 30000 32000 400 260D0 28000 200

100 34000
SECONDS SECONDS

2 5

5 VERTICAL VELOCITY ERROR DIFFERENCE

.4 a)

.3
.5

VERTICAL VELOCITY ERROR DIFFRENCE0

-2 - MAN)

MCONDS SECONDS

Figure 4-9. Velocity Differences ('TRUJTH" - OIRB)

4-23
 Table 4-4
Velocity Error State Differences

Mean Difference Ia Value Difference

Velocity Mean lo Value Mean la Value
Error States fps fps fps fps
North +0.05 0.13 -0.03 0.01
West -0.03 0.10 -0.04 0.02
Vertical +0.01 0.47 +0.06 0.05
Note: All differences are "Truth" - CIRIS

Correlated Measurement Error Analysis

The correlated measurement errors are the calibration range and range-rate biases;

X', Y', and Z' (E-frame) transponder survey errors; and the transponder atmospheric error.

These transponder errors are analyzed for the 3 transponders defined in the "Residual

Analysis" section.

Transponder Survey Errors. The Kalman filter estimates the error in the X',

Y', and Z' coordinates of the transponder survey. These error values should remain

constant in time. How well the filter estimates these errors is a function of the trajectory

of the aircraft with respect to the transponder and the ECEF axes. The values obtained

at the end of the time history are considered representative of the error magnitudes and

are presented in the following sections.

The error state means and 1 a values for transponder 181 survey errors are shown

in Figures 4-10 and 4-11. The aircraft trajectory with respect to this transponder is defined

in Figure 4-3. The aircraft trajectory completely encircles this transponder and traverses

it at different altitudes. The geometric relationship is very good and the accuracy of the

filter estimates should be high. The final X' error attained a mean of -12 ft and a 1 a
value of 2.3 ft. The Y' error attained a mean of +3 ft and a 1 a value of 3.95 ft. The

Z' error attained a mean of 0 ft and a 1 a value of 3.2 ft. The envelopes of the horizontal

4-24
 0 .1

48

4.2

FEET FEET
-1FEET
3.6 T-111 X' ERROR

3.3 (1c)
T-181 X'ERROR

(8MAN) ,
2.7

-21
2.4

2.1 1 , , II . I I I I
-24
260O 200 30000 32000 34000 26000 28000 30000 32000 34000
SECO SECON

27

24
4.8
21

18
15 T-111 VERROR 46 T-181 ERROR

(MEAN) FEET (J1)

FEET
12 4.4

4.2

I . I j . . 13.86
-3
2i0 28000 3000 32000 54000 25000 2000 30000 2000 34000
SECONI SECONDS

Figure 4-10. T-181 X and Y' Correlated Errors

4-25
 30 5

48
25

4.6
20
4.4

FEET FE T-IIIZ'(MEAN)
ERROR wET 42 T-181 Z'(10')
ERROR

10 4

5A
3.8

3.6
0

3.4

-10 , , , I , , ,, I . . 3
26000 28000 30000 3200 34000 26000 28000 3000 32000 34000

SECONDS SECONDS

30 5.1

20 48

4.5
10 T-Il ATM ERROR

FEET 39

-I0 FEET
36

-20
33

30 T-11 ATM ERROR

-40 2.7

-80 2.4 . .,
2600 22000
10 34000 26000 200 30000 000 300
SECONDS SECONDS

Figure 4-11. T-181 Z' and Atmospheric Correlated Errors

4-26
errors quickly move toward a steady state value indicating the filter is accurately deter-

mining the error magnitude. However, the Z'error does not quickly converge to a steady
value of zero. This may indicate a negligible survey error in the Z' direction or an

observability problem due to the aircraft trajectory.

The error state means and I a values for transponder 185 are shown in Figures 4-12

and 4-13. The aircraft trajectory with respect to this transponder is shown in Figure 4-3.
The aircraft tranverses the transponder only to the South of it. This geometric relationship

is the worst of the 3 transponders and the filter should have the most trouble estimating

these survey errors in all 3 ECEF axes. The final X error attained a mean of -5 ft and

a l Yvalue of 2.6 ft. The Y' error attained a mean of -4 ft and a 1 a value of 4 ft. The

Z' error attained a mean of -13 ft and a I a value of 3.5 ft. The envelopes of all 3 errors

are continuously varying and do not approach a steady state value. The accuracy of these

filter error estimates is not high, as expected.

The error state means and 1 a values for transponder 186 are shown in Figures 4-14

and 4-15. The aircraft trajectory with respect to this transponder is shown in Figure 4-3.

The aircraft transverse this transponder to the North, but it does directy overfly it. The

final X' error attained a mean of -14 ft and a 1 a value of 2.2 ft. The Y' error attained

a mean of +5 ft and a I a value of 4 ft. The Z' error attained a mean of +5 ft and a I F

value of 3 ft. The envelopes of these errors resemble the T-185 survey error envelopes,

indicating the difficulty the filter is having estimating the survey errors due to the poor

geometry.

This data indicates the transponder survey errors are observable and the filter's

random bis shaping filter model is estimating the error magnitudes. The the geomet-

ric relationship of the line-of-sight vectors (aircraft to transponder) with transponder

survey error vector determines how well the error is estimated. Therefore, the ultimate

accuracy of the survey error magnitudes is directly related to the aircraft/transponder

geometric relationship as described above.

4-27
 18 48
14
T185 X'EROR 48

12 R4EAM)
4.5

8 41
FET T5 X'ERROR

FEET
36

2
33

4
0

3
-22,
I I2.?

32000~2.4L
200 -6 30000 26000 28000 30000 32000 340D0
SECOD NCOW

0 5

49
-2
4,8

-4 4.?
J.

46 T-1/Y' RROR

-6 F 4.5

4.4

-8
4.3

-10 17415 Y'ERROR 42

4.1

-12 4
2000 30) 32O 34M0 26000 2W 30000 32000 34000
SICONiS SECONDS

Figure 4-12. T-185 X' and Y' Correlated Errors

4-28
 0 5

-2 4.8

-4 4.6

T-I Z'ERROR 4A T-IM Z'ERROR

-E (MEAN) FEET (1')
FEET 4.

-6
4

3.8

-12
3.6

-14 3.4LI . . .
260n 2W~ 30000 32M0 340 26000 2M00 300D0 3200 3CM0

ECONI SECOM

18 5.1

15 T.15 ATM ERROR

(10)
T.I ATM ERROR 4.6

12
AN) 4.5

'2

FEET FEET

-3

-6

-93

-12 2.7 I
23000 30000 32000 .4000 26000 29000 30000 32000 34000

SCONDS SECO

Figure 4-13. T-185 Z' and Atmospheric Correlated Errors

4-29
 Atmospheric Errors. The Kalman filter is estimating the line of sight error in

the range measurement due to propagation delays caused by atmospheric conditions. This

is a time varying error and is a function of weather conditions and aircraft distance from

the transponder. The range of the error and the corresponding average 1 cy value are noted

in the following sections.

The atmospheric error state plots are shown in Figures 4-11, 4-13, and 4-15 for

transponders 181, 185, and 186 respectively. This error is attributed to the changing

atmospheric index of refraction (n) which causes a decrease in the electromagnetic wave

propagation velocity, erroneously causing a increase in the measured range magnitudes.

Therefore, the filter estimated atmospheric error normally is expected to be a negative

value. However, the CIRIS computer makes an open loop correction to the range mea-

surement before it is processed by the filter. This allows the atmospheric error to vary

between negative and positive magnitudes. This characteristic is clearly observable in

the atmospheric error time histories.

The atmospheric error state mean for transponder 181 varies from -10 to +17 ft with

an average 1 Yvalue of 3.5 ft. The atmospheric error state mean for transponder 185

varies from -10 to +17 ft with an average 1 aY value of 3.5 ft. The atmospheric error state

mean for transponder 181 varies from -18 to +18 ft with an average 1 a value of 3.6 ft.

The magnitude and shape of the error time histories are reasonable for this type of error.

Therefore, this error is observable and the first-order Markov shaping filter is allowing

the filter to derive an accurate estimate for this error.

Calibration Errors. The Kalman filter is estimating the line of sight error in

range or range-rate due to a calibration error in the aircraft transponder interrogator. This

error is small and a function of the condition of the aircraft hardware.

4-30
 -2

-6 4.2

- T-I8X'ERROR .2 T-16 X'ERROR

FEET MEAN) FEET(
-10 3.6

-12 3.3

-14 3

-16 2.7

-18 2.4

-20 2.1 , ,
2m 20 3000 _0 26000 2M 300D 3
32000

SECONDS SECONDS

a S

4.8
4

2 4.6
FEET 4,-18
K T-lh Y'ERROR
(10 )
FEET
4.4
-2

-6 T-116YVIRROR
MAN)

mo0 26 ,100
MO 3m 2W 21000 30000 32000 34000
SECONDS SECON1

Figure 4-14. T-186 X' AND Y' Correlated Errors

4-31
 FEET42 T-186 Z'ERROR
-EE FEET (10')

36

T-186Z'ERROR .
-12 3GN

2800 28 35DD M20 54000 26000 28X 50000 32000 34000

SECONIM SECONDS

20 5.1

16 4.8

10 -4.5

5 42T-186 ATM ERROR

FEET FEET(1)

-5 -36

-tO 3.3
T-1I6 ATM ERROR
-15N 3

-2 ,2.7 1
200M 210M 5000 320D0 26ODD 28000 30ODC 320D0 34000

Figure 4-15. T-186 Z' and Atmospheric Correlated Errors

4-32
 The range and range-rate calibration error state means and 1 ( values are shown in

Figure 4-16. The range calibration error state mean ranges from -2.5 to +0.5 ft with an

average accuracy of 0.75 ft (1 a). The range-rate calibration error state mean ranges from

-0.13 to +0.09 fps with an accuracy of 0.006 fps (1 Y). The shape and magnitude of these

curves are reasonable for this type of error. This indicates the random bias shaping filter

model is adequate for estimating this error.

Summary
This chapter developes the "truth" model for the range and range-rate measurements.

This model is augmented to the INS "truth" model to form the full-ordered filter. The

full-orcered f ltcr is used to process 2.25 hours of empirical data from a CIRIS flight tape.

The means and 1 a values for position and velocity error states are plotted and

compared to the current CIRIS filter data. The correlated error states for the transponder

calibration, survey, and atmospheric error sources are plotted and significant character-

istics noted. Final conclusions based on this data are discussed in Chapter 6.

4-33
 1.02

.96
m

-. 6 RANGE CAL ERROR

FEET 9I'
- FEET
87

54

-2RANGE CAL ERROR .81

(MEAN)
-2.78

26000 28~3 3000

3 32000 34000 26D00 28000 30000 32000 34000

12 .1

09 .09
RANGE RATE CAL ERROR
.06 MAN).08

.07
.03

N8 RANGE RATE CAL ERROR

-.03
.04

-06
.03

-.09
02
-12
01

2M~ 2M00 3mW 32000 34CX) 2M0 2M00 30000 3200 34WO)

SECONDS SECONDS

Figure 4-16. Range and Range-Rate Correlated Errors

4-34
 V. Reduced-Order Filter Design/Simulation

This chapter develops the reduced-order filter model from the 127-state "truth"

model described in Chapter 4. The state reduction method is described and a 70-state

reduced-order filter model defined. The CIRIS flight data, processed by the filter based

on the 127-state "truth" model, is processed by the reduced-order filter based on a 70-state

model. The 127-state filter data is compared to the 70-state filter data to quantify the

change in performance. The comparison concentrates on the differences in the position

error states, velocity error states, and selected residuals.

Selection of States
This reduced-order filter is designed to function as a post processor of CIRIS flight

data. The main objective of this reduced-order filter design is to decrease the 34 hours

of CPU time (VAX 8650 computer) required to process the data from a 4.5 hour CIRIS

flight, using the 127-state filter, without a significant reduction in filter estimation

accuracy. An acceptable processing length of time is considered to be 8-12 hours (or

less) of CPU time (on a VAX 8650 computer).

The first step in eliminating states from the "truth" model is to define the states that

are not candidates for elimination. These states consist of all 42 error measurement states

(6XmmAS) defined in Appendix C; the INS (8XIs) 13 general error states (1-13), 10 cor-

related error states (14-23), and 2 baro altimeter bias error states (72,73) defined in

Appendix B. The magnitudes of these error states are significant and required to preserve

the accuracy of the basic filter model. This defines a total of 67 states as the minimum

number of states in the reduced-order filter.

The remaining 60 states (24-71,74-85) are considered based on their respective mean

and covariance error state magnitudes. The error state time history data obtained from

the 127-state filter described in Chapter 4 is used to determine these magnitudes. The

means and la values for these 60 states are plotted and compared.

5-1
 Many of these error states indicate a lack of observability because of a constant

mean and time invariant 1y value. Other states mean magnitudes are sufficiently small

to be considered insignificant. Only 3 of these 60 states have mean magnitudes which

warrant their inclusion in the reduced-order filter. The 3 gyro drift rate bias repeatability

error states (24-26) are kept. This defines the 70 error states in the reduced-order filter

model.

Filter Model

The stochastic differential equation defining the reduced-order filter's dynamics

model is shown in Eq (5-1). The primed matrices and vectors

(8X F'
s, s, , VNs, and WIS) represent a 28 error state subset of the INS matrices

and vectors defined in Eq (4-15).

8X(t) = F(X , t) 5X(t) + W(t) (5- 1)

8x = LSF_, F 0 W= LN
X(to) = [Q] P&X(to) = [Ps M A]

E { W(to)} = 0
E{ W(to)
W-o--
W T(to+,t)} = Q(to)5(T) = [Q
10
0J
Q~j 8r

The error state model for states 1 thru 26 is identical to the INS/Transponder "truth"

model defined in Chapter 4. Error states 27 and 28 in the reduced-order model are assigned

to states 72 and 73 in the INS/Transponder "truth" model. Finally, the 42 measurement

states from the INS/Transponder "truth" model (86 thru 127) are asigned to error states

29 thru 70 in the reduced-order filter model. The matrix and vector parameter/element

values are identical to the INS/Transponder "truth" model.

5-2
 The measurement model's structure for the reduced-order filter is identical to the

structure defined in Eqs (4-17) and (4-18). The only difference is the error state numbers

(29 thru 70 instead of 86 thru 127). Finally, the residual calculations are defined in Eqs

(4-19) and (4-20). The 70-state reduced-order filter is referred to as "Reduced" throughout

the remainder of this chapter.

Comparison of Reduced to "Truth"

The CIRIS flight data defined in Chapter 4 for time 26139 to 34200 is processed

with the reduced-order filter. This simulation required only 8 hours of CPU time on the

VAX 8650 computer, achieving the first part of the primary objective. The second part,

minimizing the filter estimation performance degradation, is examined next. The range

and range-rate residuals are compared first. The transponders selected for analysis in

Chapter 4 (T-181, T-185, and T-186) are again analyzed. The time histories, calculated

means, and calculated l values for the reduced-order and "truth" model filters are

compared. The differences between the two sets of residual statistics are insignificant.

Therefore, the deletion of the 57 INS error states has no significant effect on the residual

magnitudes and l values. This indicates the change between the "truth" and reduced-

order filters' position and velocity error state estimates should be small.

Another indicator of performance change is defined by inspecting the differences

between the position and velocity error states of the 127-state filter and the 70-state filter.

The filters' position and velocity error state means are differenced ("Truth" - Reduced)

and the time histories presented in Figure 5-1. The position and velocity error state Io
value differences are approximately zero for the entire time histories and are not presented.

The differences all exhibit a large negative spike at approximately time 28300. The

cause of this spike can not be attributed directly to any aspect of the aircraft trajectory,
since the flight trajectory is the same for both filters. It must be a direct result of the

deletion of a state(s) that absorbed this error in the "truth" model. This anomaly is not
considered critical and is not investigated further.

5-3
 Z 15

05

00

- DD'FERENCE
LAMUD ERROR.0 -

NORTH OCflY ERROR1 DIFFERECE

-122

2600 2800 20D0

300 34000 210X 28000 30000 32000 3400
SECONDS SECONDS
6 2

00

-2

FEET
--.

-8 LONGF E ERROR DIFFEENCE -

(MEAN) .
WEST I CT ERROR DMFERENCE
-1 -.6 (MEAN)

SECONDS SECONDS
60 7

O VETICAL VEL0CrrY E3101 DIFFEENCE

40 - ALT RERROR DIFERENCE (MEAN

20

FEET D

-20 -2

SECONDS SECONDS

Figure 5-1. Position and Velocity Differences ("TRUTH" - REDUCED)

5-4
 Generally, the differences all display a small time varying bias. The difference
grows during hard aircraft maneuvers. The small bias is directly attributed to the deletion

of the 57 INS error states in formulating the reduced-order filter design. To quantify this

difference the mean and standard deviation of the differences are calculated and presented

in Table 5-1.

Table 5-1
"Truth"/Reduced Position/Velocity Error State Differences

Mean Difference la Value

Difference
Position Mean Icy Value Velocity Mean 1ly Value
Error States ft ft Error States fps fps
Latitude +0.41 1.03 North 0.000 0.030
Longitude -0.14 0.90 West 0.000 0.050
Altitude -4.44 11.63 Vertical 0.010 0.100
Note: All differences are "Truth" - Reduced

The horizontal position error mean differences are less than a half foot with a

standard deviation of approximately 1 foot. The horizontal velocity error mean differences

are also very small. The vertical position and velocity states display the largest average

difference with a position difference on the order of 5 ft and the velocity difference of

approximately 0.01 fps.

The resulting change in performance between the "truth" and the reduced-order filter

is small. The reduction in required computer time is significant. The "truth" model based

filter required approximately 17 hours of CPU time compared to 4 hours for the

reduced-order model filter on the VAX 8650 computer system. This is a 84% reduction

in computer time with a small loss in performance.

Covariance Analysis Consideration

Ideally, the next step to complete the design of the reduced-order filter is to optimally

tune it using a covariance analysis (16:325). This consists of running the reduced-order

filter with measurements from the "truth" model and tuning the noise strengths (Q and

5-5
R) of the reduced-order filter until the the "true" error covariance is as close as possible

to the reduced-order filter's error covariance magnitude. In the previous section, only

the diaganol elements, of the error covariance matrix are compared resulting in a neglible

difference. The effects of the off-diaganol elements are ignored in this analysis.

Therefore, a covariance analysis and filter tuning can produce an increase in performance.

The MSOFE program is capable of performing this analysis. However, this analysis

requires the 127-state filter and the 70-state filter to be operated simultaneously in the

analysis. This requires the simulation of a filter on the order of 200 states.

This analysis is computationally intensive and time consuming. The comparison in

the previous section indicates only a small gain in performance. Based on these con-

siderations no attempt is made to increase the performance of the reduced-order filter

through this technique.

Comparison of the Reduced and CIRIS filters

The reduction in required computer time allows the reduced-order filter to easily

process the entire 4.5 hour flight trajectory. This simulation required only 8 hours of

CPU time on the VAX 8650 computer system. To quantify the performance of the reduced

filter, its position and velocity error state statistics are compared to the data from the

current CIRIS filter. In addition, since this complete CIRIS flight required 18 trans-

ponders, transponder switching in and out of the 10 transponder correlated error state

window is required. The performance of this technique is also presented and examined

in this section.

Position and Velocity Differences. The data position and velocity error state

data from the filter are compared to the current CIRIS filter for the entire flight. The

differences in the mean and 1y values for the entire flight (time 26139 to 42300) are

shown in Figures 5-2 and 5-3.

5-6
 40 2

300'w
A

100

20

FUEEKT -4

-20

-30
-8 AIUEERRDfUE
40 LATITUDE ERROR DIFFERECELAIUERORDFRE
1 r
-50 MN

-60 1 1 1 1 -'2 II I
2500 26000 30000 32000 34 36000 26000 40000 42000 260 28~~03000 3000 36000 ; 36000 400D0 42000

SECONDS SECONDS

ISO 0

LONGITUDE ERROR DWFEJUNCE

120 (MEAN) -
FEET FEET
90 -- 4

30LONGITUDE
ERORIFFERENCE

2@D0 28000 300 32000 34 3000363000 40000 42000 26000 28000 3DOG 32000 340 26000 36000 40000 42000

SECONDS SECONDS
SIM s0

00 ALTITUDE ERROR DIFFERENCE ALTITUDE ERROR DffnRENCE

(MAN) 40(10')
300

20
200

FEET WET
100 - C

-20

-'00

-40

-300 [
2000 28000
I
.DOD0 32000 340
I
600 000
I 1
4000
1
42000
~ -6
2650002600 3000 M20 340 300 80 00 .20
SECONDS SECONDS

Figure 5-2. Position Differences (REDUCED - CIRIS)

5-7
 N4ORTH VELOCITY ERROR DIJIERENCE

~4EA1 1 NORTH VELOCITY ERROR D(FFERENCE

.06(1O'
.2
06

.04
FPS 0FF5.0

20
-.04

26 2000 30000 32000 .54 0 60 38000 40000 42 ~ 2800 2 ;00;k0 320k 34 36000 3M00 40000 42000

SECONDS SECONDS
4 0

WEST VELOCITY ERROR DIFFEREN)CE

5 06 WEST VELOCITY ERROR DIFFEENCE

204 (1O'

02

0 "S 0

- 02

-2 -04-

- 06

-5o
260 80 200 00 0 40 300300 00 42000 26000 28000 30000 320 2400 800m60 4000 42000

SECONDS SECONDS

VERT1CAL VELOCITY ERROR DIFFERENCE VERTICAL VELOCITY ERROR DIFFERENCE

(MEAN) .6 O'

2 .4

-3 0
S 0

3r I I 1 - 1I I
2WO0 28000 300 32000 600 340
8000 40000 4200 2002~3002003000 3600 M800 40W0 4200
SECONDS SECONDS

Figure 5-3. Velocity Differences (REDUCED - CIRIS)

5-8
 The characteristics of the differences are very similar to the characteristics exhibited

in Chapter 4 between the "truth" model and the current CIRIS filter (Figures 4-8 and 4-9).

The difference time histories for flight segment not shown in Chapter 4 (time 34200 to

42300) closely resemble the segment examined in Chapter 4 (time 26139 to 34200). No

significant anomalies are identified. However, the position la value differences grow in

magnitude for this later flight portion. The cause of the error growth is not obvious, but

since the error growth is small, this anomaly is not considered critical.

To quantify the differences, the position and velocity error state differences for the

respective mean and la value differences are calculated and presented in Tables 5-2 and

5-3.
Table 5-2
Reduced/CIRIS Position Error State Differences

Mean Difference 1c Value Difference

Position Mean la Value Mean 1c Value
Error States ft ft ft ft
Latitude +1.66 10.26 -0.71 1.42
Longitude +2.98 9.56 -0.58 1.29
Altitude -15.89 83.46 +2.29 10.95
Note: All differences are Reduced - CIRIS

The position error state difference means are well with the current CIRIS accuracies,

indicating the increase in performance defined by these values as being reasonable. Since

the residual characteristics of the reduced-order filter are identical to the "truth" model

filter and both are closer to the ideal residual characteristics, the differences defined in

this section indicate the magnitude of movement of the error state estimate in the

reduced-order filter towards the true real world error.

The velocity error state differences are also similar in magnitude to mne values

calculated in Chapter 4 for the "truth" model. Again, these differences indicate a rea-

sonable movement of the filter velocity error state estimates toward the true error in the

real world.

5-9
 Correlated Error Model Performance. The correlated measurement errors
for 1 transponder are examined in this section for the entire flight. Transponder 181 is

chosen for analysis. This transponder shares it assigned correlated error states with another

transponder (transponder 163). The major thrust of this section is to examine the correlated

errors' time history for the entire CIRIS flight and to note any transponder switching

anomalies.

Table 5-3
Reduced/CIRIS Velocity Error State Differences

Mean Difference la Value Difference

Velocity Mean la Value Mean lo Value
Error States fps fps fps fps
North -0.006 0.113 -0.024 0.011
West -0.029 0.089 -0.041 0.013
Vertical +0.000 0.521 +0.048 0.065
Note: All differences are Reduced - CIRIS

The range residuals for transponders 181 and 163 are shown in Figure 5-4. The

characteristic statistics of both residuals are identical to the characteristics presented in

Chapter 4 for the "truth" model. They are shown to identify the switching times from

181 to 163 for the correlated errors.

Transponder 163 is assigned to 181 because their measurements never overlap

throughout the CIRIS flight. The 10 slot correlated measurement error window is large

enough to allow sets of non-overlaping transponders to be identified and assigned to one

of the slots for any number of transponders. The dashed lines in the figure indicate the

times when each transponder measurements are being used by the filter. The dashed lines

also indicate areas when the filter is not using either transponder.

5-10
 20
T-181 II T-163 NI T-181
15

10 E
A A,
5i

FEII " 0

-5

-15

RAN( ERESIIOUXI.
-20 L
25000 280' 3 30000 32000 34000 36000 38000 40000 42000
SECONDS
20

T-181 N T-163 -181

15
0
IM
E IE
10 A A
s! is
5

-10

20RANGE RESI L00

26000 280 0000 33200o 0 34000 36000 35000 40000 42000

SECONDS

Figure 5-4. Range Residual T-181 and T-163

The correlated measurement error time histories from the filter exhibit both the error
means and Ia values for both transponders. The correlated measurement error plots for
the X', Y', Z' transponder survey errors and the atmospheric error (ATM) in Figures 5-5

and 5-6.

Only one set of dashed lines is shown in these figure to indicate the approximate
transitions from transponder 181 to 163. The areas when neither transponder is supplying
measurements is small and lost in the transponder switching transient. The differences

5-11
 2 51
T-31 T-163 T.Ii 131 T.W 1.1

-2 45

42

~ Fl IrERROR

FEET~ FEET0.

-12 X'ERROR 27

-14 24

-16 2.1

-18I I I __ L
2M000 280 30 32000 54000 36M0 3O0 40000 42000 26000 28000 30 ,2000 34000 ,6000 380 40000 42000
SECONDS SECONI

T.131 T10 T-143 T1

4.8

12 V'FOR 4.6 Y'EOR

FEET FEET
9 4.4

4.2

AA 4

6000 2M0 MW0 32000 34000 36000 38 40000 42000 20 28M O0D 2000 34000 36000 3S0 40000 D0

SECONDS SECONDS

Figure 5-5. T-181/T-163 X' and Y' Correlated Errors

5-12
 24 .

21111T -16 4.8 1.31 14 63 1

4.8

,I (

42 (ERRO
Z'ERROR
Z'EltOl

3.6

3.3
I

2WX ?M JODDO 32000 340 000 3800 40000 42000 25000 28000 J= 32000 38000 38
WW000 400 42000

SECONDS SECONDS
2C 5.1
.- 16 T.1A1 T-181 T.10 T.-I

15 2 (10,)

4.8
0 -

ANERRO 4.5

10

-15 ATM ERROR 3

42.7

__% 4
2., l
1)000
2MCM-( 3400 3600 3-MM 42000
40DXOO 2W00 280 no00 32000 M4U0 3100 3W00 40000 4200

SECONDS SECONDS

Figure 5-6. T-181/T-163 Z' and Atmospheric Correlated Errors

5-13
in the error state means is apparent between the two transponders. The IY value changes

are less pronounced. This is due to the favorable transponder geometry when the switch

takes place. This results in a small change in the filter's confidence in its estimate. The

large transient at the beginning of flight is due to the unfavorable transponder geometry

during takeoff and this characteristic is not expected to occur during inflight switching

of transponders, as verified by these plots.

The general characteristics of the transponder survey and atmospheric errors are the

same as those shown in Chapter 4. The magnitudes are very reasonable for the type of

error being estimated. The transients when switching from 181 to 163 and back to 181

are clearly visible. The transients disappear very quickly as the filter begins tracking the

error for the new transponder. The reinitialization of the covariance matrix and mean for

these error states seems to work quite well. No anomalies are presents in this data.

Summary
In this chapter the reduced-order filter design is presented. The reasons for deleting

the 57 INS error states are explained and the reduced-order filter model defined. The

performance change is quantified by comparing the residuals, position error states, and

velocity error states statistics between the "truth" and reduced-order filters.

A covariance analysis is considered for further tuning of the reduced-order filter and

the reasons for rejecting this option explained. The reduced-order filter's performance is

compared to the current CIRIS filter to quantify the performance gains. Finally, the

correlated measurement errors for 2 transponders are examined for the entire CIRIS flight.

Special attention is given to the transponder switching effects on the correlated mea-

surement error mean and la value time histories.

5-14
 VI. Conclusions and Recommendations

This chapter restates the major goals of this research and the level of attainment for

each. This analysis is integrated into the conclusions section that follows. Based on the

results of this research, areas for future exploration and improvement are defined in the

recommendations section.

Conclusions

The primary goal of this research effort is to develop a full ordered "truth" model

which represents the errors found in the real world as close as possible. This goal is

attained with both the 127-state and 70-state filters developed. This claim is based on

how well the filter range and range rate residual statistics resemble the ideal residual

statistics and the comparison to the current CIRIS filter error estimates and residuals.

The residual statistics for the 127-state and the 70-state filter are nearly identical.

The residual statistics are very close to zero mean and the filter predicted lo value,

indicating that the filter models are correctly estimating the real world errors. The current

CIRIS residuals are found to be biased and with clearly visible correlated errors. The

current CIRIS residuals' Ia values are significantly different from the predicted filter la

values. This indicates the position and velocity error state estimates for the 127-state

and 70-state filters are closer to the true error.

To quantify the increase in performance the differences between the "truth"/reduced

order filter and current CIRIS filter, position and error state means are calculated and

presented in Tables 6-1 and 6-2. The differences between the filter predicted la values

are not significant and not presented here.

6-1
 The horizontal position error increase in accuracy is on the order of 2 ft for both

the 127-state and 70-state filters. The vertical position error increase in accuracy in on

the order of 20 ft for both filters. This increase in accuracy is well within the current

CIRIS independently verified position accuracies. Therefore, the values obtained are very

reasonable.

Table 6-1
"Truth"/Reduced/CIRIS Position Error State Differences

Mean Difference Mean Difference

("Truth" - CIRIS) (Reduced - CIRIS)
Position Mean la Value Mean lo Value
Error States ft ft ft ft
Latitude +2.11 10.55 +1.66 10.26
Longitude +2.29 9.43 +2.98 9.56
Altitude -20.61 83.49 -15.89 83.46
Note: All differences are Reducedf'Truth" - CIRIS

The North velocity error increase in estimation accuracy is on the order of 0.005
fps for both the 127-state and 70-state filters. The West velocity error in estimation
accuracy is 0.03 fps for both filters. The vertical velocity error increase in estimation

accuracy is negligible for both filters. This increase in accuracy is well within the current
CIRIS independently verified velocity accuracies. Therefore, the values obtained are

reasonable.
The correlated measurement survey and atmospheric errors are observable to the
filter and the filter is attempting to estimate these errors. The error mean and la values

magnitudes seem reasonable for this type of error. At this time, there is no precise method
of deterrr;ning the accuracy of these estimates. However, based on the "whiteness" of
the filter residuals, these estimates most likely are close to the real world errors.

The performance differences between the 127-state and 70-state filter are very small.

However, the difference in required computer time is significant. Either model could

serve as a "truth" model in the development of a Kalman filter that includes aiding
measurements from the GPS system.

6-2
 Table 6-2
"Truth"/Reduced/CIRIS Velocity Error State Differences

Mean Difference Mean Difference

("Truth" - CIRIS) (Reduced - CIRIS)
Velocity Mean la Value Mean 1c Value
Error States fps fps fps fps
North -0.002 0.105 -0.006 0.113
West -0.029 0.104 -0.029 0.089
Vertical +0.005 0.522 +0.000 0.521
Note: All differences are "Truth"/Reduced - CIRIS

Recommendations
Additional CIRIS Flights. All the research in this thesis is based on the data
from a single CIRIS flight. In order to validate this model, data from different CIRIS

flights, with varying flight dynamics, should be processed with the filter. Ideally, a

comparison of this data with data from a higher accuracy system would definitely define

the true increase in accuracy. Only a differential GPS system or some other ground test

system such as the high speed test track, can provide sufficiently accurate reference.

Correlated Errors. The correlated measurement model for the survey and

atmospheric errors is providing estimates that appear reasonable for this type of error.

Based on the data from more flights, these models may be tuned (initial conditions,

correlation times, and noise strengths) to improve performance.

Filter Tuning. The "truth" models parameters (such as noise strengths, corre-

lation times, and initial conditions) could be adjusted, using numerous CIRIS flights, to

draw the residual statistics closer to the ideal statistics. The remaining increase in

performance may be limited by hardware limitations (INS and RRS). Therefore the

expected performance increase remaining through the tuning of this "truth" model is small.

6-3
 The reduced order filter could be tuned using a covariance analysis to move the

performance of the reduced order filter estimates closer to the "truth" model. Again, the

differences between the "truth" and reduced order filters is small and this tuning may not

be worth the effort.

Acceleration Data. The acceleration data directly available from the LN-39 INS

is not being used to drive the filter model. Instead, the acceleration data is computed

from the INS velocity data. There may be some benefits to using the raw INS acceleration

data, especially in high flight dynamic environment. The benefit, if any, is unknown at

this point.

Higher Sampling Rate. Currently measurement and INS data is processed at

a 1 Hz rate due to computer limitations. This sampling rate seems adequate for the benign

flight envelopes of the large cargo aircraft. However, when the CIRIS system is used in

high dynamic fighters a higher sampling rate may allow the filter to accurately estimate

the rapidly changing INS position and velocity errors. At this time the expected increase

in performance cannot be estimated.

Smoothing Algorithm. The most promising way to increase the accuracy of

the 127-state or 70-state filter is to implement the Kalman filter using a smoother algo-

rithm. The smoothing algorithm utilizes a forward and backward Kalman filter to estimate

the error states (17:1-5). This algorithm can only be used in post processing, but it takes

into consideration the measurements for the entire flight when it makes its error estimates.

This algorithm would greatly reduce the areas of poor accuracy due to degraded

transponder coverage. It would also do a better job estimating the transponder survey

errors. This research direction will obtain the best results in the shortest time.

6-4
 GPS Measurements. The 127-state or 70-state filters designed in this research

effort more accurately estimate the INS position and velocity errors as is evident in a

comparison of the range and range rate residuals. Therefore, either model can serve as

a basis for the GPS aided CIRIS system. The filters designed in this thesis have the added

benefit of providing a check on the transponder survey locations. If any of the transponder

survey points is in error it would be flagged and the error estimated in the correlated

survey error states.

The 70-state filter is recommended for the basis of GPS research. This is because
of the large decrease in computer time and a negligible decrease in performance (as
compared to the 127-state filter). The GPS research should be given higher priority over
the tuning of the INS/Transponder filter. The only significant gains in accuracy, con-
vergence, and reliability will be achieved when GPS measurements are added to the model.

Initial work in this area is presented in Reference (20).

6-5
 Appendix A
A Historical Review of the CIRIS Development and Operation

This appendix presents a brief review of the major characteristics of the CIRIS

development and operation. This appendix is taken directly from References (11) and

(12).

A-i
Introduction
The Completely Integrated Reference Instrumentation System (CIRIS) provides a

highly accurate, real-time, position, velocity, and attitude reference for flight tests of
inertial navigation, guidance, and radar systems. CIRIS was developed as an in-house
project by the Central Inertial Guidance Test Facility (CIGTF) of Holloman AFB, New

Mexico in 1975.
This airborne automated system is configured to be carried aboard both cargo and
fighter testbed aircraft, and is also used in a mobile testing van. Utilizing ground-based
transponders positioned in a triangular pattern along the flight path of the aircraft, the

CIRIS provides reference coverage throughout most of the continental Unites States. This
enables the testing of inertial systems over a very long distance and time duration. Position
accuracies are provided at 13 ft (l) horizontal and 40 ft (1a) vertical. Velocity accuracies
are provided at 0.1 fps (1 Y)horizontal and 0.4 fps (1a) vertical. It provides the Department

of Defense with a valuable airborne test facility for use in development and verification

of inertial systems.

Past Reference Systems Compared to CIRIS

Air Force flight testing of inertial navigation systems began in 1965 at Holloman
AFB, New Mexico. From 1965 to 1975 four primary methods were used by CIGTF in
generating reference data: checkpoint, radar, cinetheodolite, and Doppler velocity. The

checkpoint method utilizes an airborne vertical camera mounted in the testbed aircraft.
High resolution photographs are taken of precisely surveyed landmarks. Post flight
evaluation of the photographs, correlated with time and test system data, enabled the

analyst to determine system errors. In the radar method, radar facilities are used to track
the aircraft during a test, the data is correlated with the system data for evaluation. In
the cinetheodolite method, cinetheodolite instrumentation (CINE) enables ground-based

A-2
precision-pointed cameras to track the aircraft. Picture analysis determines aircraft

position and velocity. The Doppler Velocity and Positionary System (DOVAP) measures

the incremental position changes based on the frequency shift of a signal between the
aircraft transponder and ground-based receivers. Table A-I summarizes the characteristics
of these four reference systems and compares them with CIRIS.

Table A- I
Reference System Comparison

Checkpoint Radar CINE DOVAP CIRIS

Position (ft) 50-200 50-100 5-15 15 13
Velocity (fps) ---- 10-20 0.5-1.0 0.5 0.1
Attitude (arcmin) ---- ---- ------ ----- 6
Data Available 2-3 7 21 28 Real
(days) Time
Personnel Reqd 2 10 10 20 1
Weather Dep. Yes No Yes No No
Coverage (mins) -
North-South" > 84"" 42 6 20 > 84
East-West > 84 84 2 5 > 84
* Assume constant track starting at Holloman AFB, NM, 300-400 Knots
groundspeed.

** 84 minute is defined as the Schuler period. Test durations are measured in incre-
ments of Schuler periods.

Development of CIRIS
The data presented in Table 1 indicates the reference systems, used prior to CIRIS,

involved long time data turnaround, limited the range of testing, and often used more

support personnel. These facts combined with advances in digital techniques, specifically

Kalman filtering, led to the conceptualization of a self-contained, real-time airbome

reference system. The CIRIS system was developed as an in-house project of CIGTF.

The CIRIS development effort was completed in March 1975 with four verification flight

A-3
tests conducted over the White Sands Missile Range. Dense transponder coverage
combined with extensive CINE data verified the CIRIS position accuracy. Since this time
CIRIS has been reconfigured for fighter and B52 aircraft, and is used heavily for inertial
system's flight testing.

CIRIS in an Operational Flight Test

CIRIS is configured for installation in NCI30, NC141, B52, and RF-4C testbed
aircraft. On board cargo aircraft CIRIS is installed on a pallet along with four test system
pallets. In the RF-4C the CIRIS is modified for containment in a pod secured to the
aircraft. Two test systems are carried on board the F-4. All test systems are time syn-
chronized with CIRIS for data correlation. The actual flight is conducted over areas where
CIRIS transponders are positioned in a triangular pattern, 100-200 miles apart. This
reference system provides steering information to the aircrew to fly precise ground tracks.
Transponder ranges are available at many different locations throughout the United States
land mass. CIRIS personnel are also able to reposition ground based transponders to
provide coverage in a specific area of test activity. New transponder sites need only be
surveyed prior to placement of the transponder units. Routes and area can change subject
to test activity in that region.

CIRIS System Description

The major components of CIRIS are the Inertial Navigation System (INS), the Air
Data Computer (ADC), the Radio-Range/Range Rate/Interrogator System (RRS), and the

time generator interface with the CIRIS computer. There are currently two versions of
CIRIS being used designated CIRIS I and CIRIS II.
The Litton LN-15 INS, a Sperry ADC, the Cubic RRS, and a 11-state Kalman filter

(initially designed by Intermetrics) are used in CIRIS I to combine all the available data
to produce the most accurate aircraft trajectory data possible. The Sperry ADC provides

A-4
barometric altitude to the INS for vertical channel stabilization. The Kalman filter
combines the INS position, velocity, and acceleration data; the barometric altimeter data;
and the very accurate line-of-sight range and range rate measurements from the Cubic
RRS to estimate the error in the INS position and velocity data. These Kalman filter

estimated errors are used to correct the INS data trajectory data. The CRIS filter algorithm
is implemented on a Hewlet Packard HP2100 minicomputer.

A mass producible high reliability and maintainability improvement to the previous

CIRIS I design resulted in an improved system designated CIRIS II. This system was
developed and fabricated by the Navigational Reference Branch (GDN). Its old INS is
replaced with the newer, depot supported LN-39 INS. It also removes the non-standard
computer interface to the old INS and replaces a non-available ADC with a depot sup-
ported F-16 ADC. The accuracy of CIRIS I and CIRIS H1 are identical, only the reliability
and maintainability of the CIRIS II system is increased.

CIRIS Applications
With its ability to be used anywhere, dependent only upon positioning of trans-

ponders, CIRIS has been heavily used in the past. Flight testing of inertial navigation
systems for the A-10, B-52, F-15, F-16, B-l, and KC135 have been accomplished using

CIRIS as the reference. A GPS system was tested onboard a C141 with CIRIS providing
reference data. Currently CIRIS is extensively used in testing strapdown systems by

CIGTF, as well as being used in the B52 upgrade (B52-OAS Program). Evaluation of
mid course and terminal guidance systems use CIRIS not only as the reference, but as a
source of data for in flight alignment prior to simulated weapons delivery.
The fighter CIRIS is completely automated thus insuring optimum reference data

in high g-force fighter tactics environment. Plans to miniaturize the CIRIS call for
installation of microcomputer and cassette type recording system. The fighter CIRIS is

A-5
being configured for the F-15B testbed aircraft and more efficient cooling for the fighter

unit is also in research. Placement of more transponder units is increasing range and

capability of CIRIS. CIRIS engineers are constantly modifying software to meet

requirements for both standard and special test programs. Plans for GPS update of CIRIS

in minimal transponder coverage are also seen in the CIRIS future. Research sponsored

by CIGTF is conducted by the Air Force Institute of Technology (AFIT) to quantify the

gains in filter error estimation accuracy possible by adding GPS aiding measurements to

CIRIS.

In summary, CIRIS is a valuable tool in the testing of inertial systems. The highly

accurate, real-time reference data provides the Air Force and Department of Defense with

a standard by which commercially built inertial systems can be verified and new systems

can be developed and tested.

A-6
 Appendix B
Initial Covariance, Noise Strengths. and Dynamics Matrix Elements

for the INS "Truth" Model

This appendix defines the elements of the initial covariance matrix for the INS

truth" model. The diagonal matrix PIs(to) is shown in Eq (3-6). The submatrices P,,

thru P,, are also diagonal matrices. The I Yerror values for each diagonal element are

defined in Tables B-I thru B-9. Note that the actual covariance diagonal element is equal

to a;.
The elements of the initial noise strength matrix are also defined. The block diagonal

matrix, Qus(to), is shown in Eq (3-7). The nonzero block diagonal elements (Q,, and

Q21) are also diagonal matrices. The diagonal element values are defined in Tables B-10

thru B-I l with the applicable parameters defined in Chapter 3.

Finally, the nonzero elements of the FNs matrix are defined. The structure of this

matrix is shown in Eq (3-3). The submatrices F,, thru F6 are defined in Tables B-12

thru B-24 with the applicable parameters defined in Chapter 3.

B-i
 Table B-I
Definition of Submatrix PI1

Element State Definition Value

____ _ __ ___(1 a)
1,1 S3x Magnitude of error between the true frame's X axis 0.0
and the computer frame's X axis. arcsecs

2,2 8ey Magnitude of error between the true frame's Y axis 0.0
and the computer frame's Y axis. arcsecs

3,3 80z Magnitude of error between the true frame's Z axis 0.0
and the computer frame's Z axis. arcsecs

4,4 x Magnitude of error between the true frame's X axis 100

and the platform frame's X axis. arcsecs

5,5 Or Magnitude of error between the true frame's Y axis 100

and the platform frame's Y axis. arcsecs

6,6 Oz Magnitude of error between the true frame's Z axis 180

and the platform frame's Z axis. arcmins

7,7 8Vx Magnitude of error between the true X velocity and 0 fps
the computer X velocity.

8,8 8VY Magnitude of error between the true Y velocity and 0 fps
the computer Y velocity.

9,9 8Vz Magnitude of error between the true Z velocity and 0 fps
the computer Z velocity.
10,10 SH Magnitude of error between the true altitude and the 0 ft
computer altitude.

11,11 &q L Magnitude of error between the true altitude and the 0 ft
computer altitude (1 sec delay).
12,12 &5 3 Magnitude of vertical channel aiding state. 0 ft
13,13 &54 Magnitude of vertical channel aiding state. 0 fps 2

All other elements are zero.

B-2
 Table B-2
Definition of Submatrix P22

Element State Definition Value

14,14 Bxc Magnitude of gyro correlated drift rate along X axis. 0.002
deg/hr
15,15 Byc Magnitude of gyro correlated drift rate along Y axis. 0.002
deg/hr

16,16 Bzc Magnitude of gyro correlated drift rate along Z axis. 0.005
deg/hr

17,17 Axe Magnitude of accelerometer and velocity quantizer 3

correlated noise along X axis. 9ig

18,18 Arc Magnitude of accelerometer and velocity quantizer 3

correlated noise along Y axis. jig

19,19 Azc Magnitude of accelerometer and velocity quantizer 3

correlated noise along Z axis. ig

20,20 5G x Magnitude of gravity vector error along X axis. 35

jig
21,21 8Gy Magnitude of gravity vector error along Y axis. 35
ig
22,22 8G z Magnitude of gravity vector error along Z axis. 35
9ig

23,23 al-Ic Magnitude of baro-altimeter correlated bias noise. 100

ft

All other elements are zero.

B-3
 Table B-3
Definition of Submatrix P33.

Element State Definition Value

(I ;a
)
24,24 Bx Magnitude of gyro drift rate bias repeatability along 0.0045
X axis. deg/hr
25,25 By Magnitude of gyro drift rate bias repeatability along 0.0045
Y axis. deg/hr
26,26 Bz Magnitude of gyro drift rate bias repeatability along 0.01
Z axis. deg/hr
27,27 S Magnitude of gyro torquing scale factor error along 0.021
X axis. %
28,28 SGY Magnitude of gyro torquing scale factor error along 0.021
Y axis. %
29,29 S Magnitude of gyro torquing scale factor error along 0.021
Z axis. %
30,30 X1 Magnitude of gyro misalignment about Y axis. 7
arcsec
31,31 X2 Magnitude of gyro misalignment about X axis. 7
arcsec
32,32 X3 Magnitude of gyro misalignment about X axis. 7
arcsec
33,33 N, Magnitude of gyro misalignment about Z axis. 7
arcsec
34,34 N2 Magnitude of gyro misalignment about Z axis. 7
arcsec
35,35 N3 Magnitude of gyro misalignment about Y axis. 7
arcsec

B-4
 Table B-4
Definition of Submatrix P33b

Element State Definition Value

______
__ ~~ _ ~
__ _ ~
_F__ ~ ~
__ ~
___ _ _ ~
_l) (

36,36 Hx Magnitude of error due to gyro mass imbalance 0.0 16

_______ _______along X axis (g-sensitivity). deg/hr/g

37,37 Hy Magnitude of error due to gyro mass imbalance 0.016

________along Y axis (g-sensitivity). deg/hr/g

38,38 Hz Magnitude of error due to gyro mass imbalance 0.0 16

________along Z axis (g-sensitivity). deg/hr/g

39,39 Hy Magnitude of error due to gyro quadrature along X 0.0 15

axis (g-sensitivity). deg/hr/g
40,40 Hx Magnitude of error due to gyro quadrature, along Y 0.015
axis (g-sensitivity). deg/hr/g
41,41 Hy Magnitude of error due to gyro quadrature along Z 0.015
_______________axis (g-sensitivity). deg/hr/g

42,42 Kz Magnitude of error due to gyro anisoelasticity along 0.02

X axis &g-sensitivity). deg/hr/g 2

43,43 Kz Magnitude of error due to gyro anisoelasticity along 0.02

Y axis &g-sensitivity). degfhr/g2
44,44 Kx Magnitude of error due to gyro anisoelasticity along 0.02
2 -sensitivity). 2
______Z axis (g Ideg/hr/g

All other elements are zero.

B- 5
 Table B-5
Definition of Submatrix P4.

Element State Definition Value

(I)
45,45 Ax Magnitude of accelerometer bias repeatability 200
along X axis. Ig
46,46 A1 r Magnitude of accelerometer bias repeatability 200
along Y axis. pg
47,47 ABz Magnitude of accelerometer bias repeatability 200
along Z axis. 9g

48,48 SAX Magnitude of accelerometer and velocity quan- 0.023

tizer scale factor error along X axis. %

49,49 SAY Magnitude of accelerometer and velocity quan- 0.023

tizer scale factor error along Y axis. %

50,50 SA, Magnitude of accelerometer and velocity quan- 0.023

tizer scale factor error along Z axis. %

51,51 SQAx Magnitude of accelerometer and velocity quan- 0.01

tizer scale factor asymmetry error along X axis. %

52,52 SQA, Magnitude of accelerometer and velocity quan- 0.01

tizer scale factor asymmetry error along Y axis. %

53,53 SQA4 Magnitude of accelerometer and velocity quan- 0.01

tizer scale factor asymmetry error along Z axis. %
Magnitude of accelerometer and velocity nonlin- 3
54,54 SQX ear quantizer scale factor asymmetry error along g g/g 2
X axis.
Magnitude of accelerometer and velocity nonlin- 3
55,55 SQ,2 ear quantizer scale factor asymmetry error along t g/g 2
Y axis.
Magnitude of accelerometer and velocity nonlin- 3
56,56 S Z2 ear quantizer scale factor asymmetry error along R.g/g 2
Z axis.

B-6
 Table B-6
Definition of Submatrix PUb

Element State Definition Value

(1 a)
57,57 Fxx Magnitude of error proportional to the square of the 8
measured acceleration along X axis. i g/g 2

58,58 Fyy Magnitude of error proportional to the square of the 8

measured acceleration along Y axis. i g/g 2

59,59 Fzz Magnitude of error proportional to the square of the 8

measured acceleration along Z axis. i g/g 2

Magnitude of error proportional to products of accel- 33

60,60 Fxr eration along and orthoganal to sensitives axes of i g/g 2
accelerometers (XZ).
Magnitude of error proportional to products of accel- 33
61,61 Fxz eration along and orthoganal to sensitives axes of i g/g 2
accelerometers (XY).
Magnitude of error proportional to products of accel- 33
62,62 Frx eration along and orthoganal to sensitives axes of i g/g 2
accelerometers (YZ).
Magnitude of error proportional to products of accel- 33 2
63,63 Frz eration along and orthoganal to sensitives axes of g/g
RX
accelerometers (YX).
Magnitude of error proportional to products of accel- 33
64,64 Fzv eration along and orthoganal to sensitives axes of g.g/g 2
accelerometers (ZX).
Magnitude of error proportional to products of accel- 33
65,65 Fzy eration along and orthoganal to sensitives axes of R.g/g 2
accelerometers (ZY).

B-7
 Table B-7
Definition of Submatrix P44,

Element State Definition Value

1(F)
66,66 RI Magnitude of error due to accelerometer misalign- 7
ment along X axis (mean). arcsec
67,67 P Magnitude of error due to accelerometer misalign- 7
ment along Y axis (mean). arcsec
68,68 R Magnitude of error due to accelerometer misalign- 7
ment along Z axis (mean). arcsec

69,69 0, Magnitude of error due to accelerometer misalign- 7

ment along X axis (1 sigma). arcsec

70,70 02 Magnitude of error due to accelerometer misalign- 7

ment along Y axis (1 sigma). arcsec

71,71 03 Magnitude of error due to accelerometer misalign- 7

ment along Z axis (1 sigma). arcsec
72,72 6H1 Magnitude of baro-altimeter bias error. 300
ft

73,73 SHs& Magnitude of baro-altimeter scale factor error. 4

1 %

All other elements are zero.

B-8
 Table B-8
Definition of Submatrix Ps

Element State Definition Value

(T)
74,74 AxQ Magnitude of initial accelerometer bias thermal tran- 200
sient error along X axis. 9g
75,75 AYQ Magnitude of initial accelerometer bias thermal tran- 200
sient error along Y axis. Rig
76,76 AzQ Magnitude of initial accelerometer bias thermal tran- 200
sient error along Z axis. 9ig
77,77 Bxo1 Magnitude of initial gyro drift rate bias thermal 0.33
transient along X axis (component 1). deg/hr
78,78 BXQ 2 Magnitude of initial gyro drift rate bias thermal 0.01
transient along X axis (component 2). deg/hr
79,79 BY1 1 Magnitude of initial gyro drift rate bias thermal 0.33
transient along Y axis (component 1). deg/hr
80,80 BYQ 2 Magnitude of initial gyro drift rate bias thermal 0.01
transient along Y axis. (component 2). deg/hr
81,81 BzQI Magnitude of initial gyro drift rate bias thermal 0.33
transient along Z axis (component 1). deg/hr
82,82 Bz2 2 Magnitude of initial gyro drift rate bias thermal 0.01
transient along Z axis (component 2). deg/hr
All other elements are zero.

B-9
 Table B-9
Definition of Submatrix P6

Element State Definition Value

__(1 )

83,83 BxrA Magnitude of gyro turn-around drift rate along X 0.002

axis. deg/hr

84,84 BrrA Magnitude of gyro turn-around drift rate along X 0.002

axis. deg/hr

85,85 BzA Magnitude of gyro turn-around drift rate along X 0.005

axis. deg/hr
All other elements are zero.

B- 10
 Table B-IO
Definition of Submatrix Q1

Element State Definition Value

4,4 Ox Magnitude of gyro drift rate white noise along
X axis.

5,5 oy Magnitude of gyro drift rate white noise along,° 2

Y axis.
6,6 Oz Magnitude of gyro drift rate white noise alonga 2
Z axis.
7,7 8Vx Magnitude of accelerometer white noise along a2
X axis. '9Aj

8,8 8VY Magnitude of accelerometer white noise along,(°2

Y axis. X

9,9 5Vz Magnitude of accelerometer white noise along a2

Z axis. '9Ax

13,13 &S4 Magnitude of vertical aiding channel white 2 K428utc"

-,
noise component.
All other elements are zero.

B-11
 Table B-I I
Definition of Submatrix Q22

Element State Definition Value

14,14 Bxc Magnitude of white noise along X axis. 2
2PBXc~r~xc

15,15 Bzc Magnitude of white noise along Y axis.

2

16,16 Azc Magnitude of white noise along Z axis.

2P .cxz

17,17 Akc Magnitude of white noise along X axis.

2PAxcO.2xc

18,18 Ac Magnitude of white noise along Y axis. 2

19,19 A Magnitude of white noise along Z axis.

20,20 Gx Magnitude of white noise along X axis. 20&O x

21,21 G y Magnitude of white noise along Y axis. 2ptIWY

22,22 Wz Magnitude of white noise along Z axis. 2 p8Go.B'G,

23,23 &Hc Magnitude of baro-altimeter correlated bias

white noise.
All other elements are zero.

B-12
 4 N4

el en

CNC

- 4 e n r- \ 00 al r4 e
 III

i-

N 9

00 -n
400

en0000 O
 x-

oiN
cqZ0.

IiS
0

rn-

ov
 SN

Lto

- 00

00

ol9

9m

00 en W nr 0 c

14, • " 9 0 - '

 N

~OLN

40 N

r-
W-

04 V)% -0 7
oo
r-4

000

C4 \ r-
(nW) 0 47SC4 e
 J9

- 9q

° 0 N,-

9m

-. " 94 t O " r
 0

m ~~ %n0-C4 e
- kU

en

ac

C1 f
-q 14 C4
00N

o00

000

000 00
oo0

00

00 00 00
 Appendix C
Initial Covariance, Noise Strengths. Dynamics Matrix,

and Measurement Gradient Matrix Elements

for the Measurement "Truth" Model

This appendix defines the elements of the initial covariance matrix for the mea-
surement "truth" model. PmE.s(to) is a diagonal matrix and is shown in Eq (4-15). The
submatrices PmFl. thru PMAs, are also diagonal matrices. The 1 a error values for each
diagonal element are defined in Tables C-1 thru C-4. Note that the actual covariance

element is equal to a2 .
The elements of the initial noise strength matrix are also defmed. QMEs(t.) is a

diagonal matrix and is shown in Eq (4-15). The nonzero diagonal element values are
defined in Table C-5.
The nonzero elements of the FMES matrix are defined. The nonzero elements are
defined in Table C-6.
The nonzero elements of the FR and HR matrices are defined. The nonzero elements
are defined in Tables C-7 and C-8.

C-I
 Table C-I
Definition of Submatrix P s.

Element State Definition Value

(Ia)

86,86 8EcAL Magnitude of the transponder interrogator calibration I

error (range measurement). ft

87,87 8XT, Magnitude of transponder 1 survey error (X'-axis). 5

ft

88,88 8YT, Magnitude of transponder I survey error (Y'-axis). 5

ft

89,89 87_4 Magnitude of transponder I survey error (Z'-axis). 5

ft

90,90 8EATM, Magnitude of transponder 1 atmospheric error (line 5

of sight). ft

91,91 8XT2 Magnitude of transponder 2 survey error (X'-axis). 5

ft

92,92 8YT2 Magnitude of transponder 2 survey error (Y'-axis). 5

ft

93,93 8r2 Magnitude of transponder 2 survey error (Z'-axis). 5

ft

94,94 8EATM2 Magnitude of transponder 2 atmospheric error (line 5

of sight). ft

95,95 8XT, Magnitude of transponder 3 survey error (X'-axis). 5

ft

96,96 8YT Magnitude of transponder 3 survey error (Y'-axis). 5

ft

97,97 843 Magnitude of transponder 3 survey error (Z'-axis). 5

ft

98,98 SEAT, Magnitude of transponder 3 atmospheric error (line 5

of sight). ft

C-2
 Table C-2
Definition of Submatrix PMEASb

Element State Definition Value

0(1)

99,99 8X-r Magnitude of transponder 4 survey error (X'-axis). 5

ft

100,100 SYT, Magnitude of transponder 4 survey error (Y'-axis). 5

ft

101,101 57_,, Magnitude of transponder 4 survey error (Z'-axis). 5

ft

102,102 SEA-M, Magnitude of transponder 4 atmospheric error (line 5

of sight). ft

103,103 5XT, Magnitude of transponder 5 survey error (X'-axis). 5

ft

104,104 SYT, Magnitude of transponder 5 survey error (Y'-axis). 5

ft

105,105 6Zr, Magnitude of transponder 5 survey error (Z'-axis). 5

ft

106,106 8EAm, Magnitude of transponder 5 atmospheric error (line 5

of sight). ft

107,107 XT6 Magnitude of transponder 6 survey error (X'-axis). 5

ft

108,108 sYT, Magnitude of transponder 6 survey error (Y'-axis). 5

ft

109,109 7-r, Magnitude of transponder 6 survey error (Z'-axis). 5

ft

110,110 8EAN Magnitude of transponder 6 atmospheric error (line 5

of sight). ft

C-3
 Table C-3
Definition of Submatrix Pmm~s

Element State Definition Value

0()

111,111 8XT, Magnitude of transponder 7 survey error (X'-axis). 5

ft
5
112,112 YT, Magnitude of transponder 7 survey error (Y'-axis). 5
ft

113,113 Zr, Magnitude of transponder 7 survey error (Z'-axis). 5

ft

114,114 8EATI Magnitude of transponder 7 atmospheric error (line 5

of sight). ft

115,115 8XT, Magnitude of transponder 8 survey error (X'-axis). 5

ft

116,116 8YT, Magnitude of transponder 8 survey error (Y'-axis). 5

ft

117,117 &Zr. Magnitude of transponder 8 survey error (Z'-axis). 5

ft

118,118 8EATMS Magnitude of transponder 8 atmospheric error (line 5

of sight). ft

8
119,119 XT Magnitude of transponder 9 survey error (X'-axis). 5
ft

120,120 8YT9 Magnitude of transponder 9 survey error (Y'-axis). 5

ft

121,121 87_ Magnitude of transponder 9 survey error (Z'-axis). 5

ft

122,122 8EATm Magnitude of transponder 9 atmospheric error (line 5

of sight). ft

C-4
 Table C-4
Definition of Submatrix P,.Ad

Element State Definition Value

0 ()

124,124 'XTI ° Magnitude of transponder 10 survey error (X'-axis). 5

ft

125,125 8YTIo Magnitude of transponder 10 survey error (Y'-axis). 5

ft
126,126 87r , Magnitude of transponder 10 survey error (Z'-axis). 5
ft
8
126,126 EA.-M, ° Magnitude of transponder 10 atmospheric error (line 5
of sight). ft
127,127 BEcAtaR Magnitude of the transponder interrogator calibration .01
error (range-rate measurement). ft
All other elements are zero.

C-5
 Table C-5
Definition of Submatrix QmRs

Element State Definition Value

90,90 BEATMI Magnitude of transponder 1 atmospheric error 0.1667

white noise strength. ft2/sec

94,94 SE^m Magnitude of transponder 2 atmospheric error 0.1667

white noise strength. ft2 /sec

98,98 8EAT Magnitude of transponder 3 atmospheric error 0.1667

white noise strength. ft2/sec

102,102 SEATU Magnitude of transponder 4 atmospheric error 0.1667

white noise strength. ft2/sec

106,106 SEA~T, Magnitude of transponder 5 atmospheric error 0.1667

white noise strength. ft/sec

110,110 SEA Magnitude of transponder 6 atmospheric error 0.1667

white noise strength. ft2/sec

114,114 8EA , Magnitude of transponder 7 atmospheric error 0.1667

white noise strength. ft/sec
118,118 8E,,T. Magnitude of transponder 8 atmospheric error 0.1667
white noise strength. ft/sec

122,122 8EA~TM Magnitude of transponder 9 atmospheric error 0.1667

white noise strength. ft2/sec

126,126 8EAr, 0, Magnitude of transponder 10 atmospheric error 0.1667

white noise strength. ft2/sec
All other elements are zero.

C-6
 Table C-6
Definition of Submatrix Fz

Element State Defirition Value

90,90 6ETM, Magnitude of transponder 1 atmospheric error -1/300
time constant. sec'I

94,94 8E^M Magnitude of transponder 2 atmospheric error -1/300

sec 1
time constant.

98,98 SEATM3 Magnitude of transponder 3 atmospheric error -1/300

time constant. sec t

102,102 8EATM, Magnitude of transponder 4 atmospheric error -1/300

time constant. sec t

106,106 8EAM Magnitude of transponder 5 atmospheric error -1/300

time constant. sec "'

110,110 8EAT Magnitude of transponder 6 atmospheric error -1/300

time constant. sec t

114,114 EAxT, Magnitude of transponder 7 atmospheric error -1/300

time constant. sec'

118,118 8EATh Magnitude of transponder 8 atmospheric error -1/300

time constant. sec'

122,122 8EAT, Magnitude of transponder 9 atmospheric error -1/300

time constant. sec'

126,126 8EATo Magnitude of transponder 10 atmospheric error -1/300

time constant. sec t
All other elements are zero.

C-7
 Table C-7
Definition of matrix HR

Element State Definition

,l 8 H,
1,2 s0y H2

1,10 8H H3

1,29 8ECL -1

8
1'#Yi XT + XT - XA

+
I,#Yi s(Tj + YT - YA

1,#Zi 8zr, +Zr- ZA

1 ,AIM 8EAT +

All other elements are zero.

[H; H; H =
fr T - f( A 'r - '(lA 2--, - ^A C (L, X,, [TX& ' , ) (C-I,

Notes: 1. i = 1,...,10 with element number selected from Tables C-1 thru C-4.
2. (XT, IT 7r) corresponds to transponder mapped to the ith element.

3. H' transformation is required to transform from INS position frame to

ECEFframe.
4. CE is defined in Eq (2-6).
5. T, is defined in Eq (3-29).
6. The # symbol represents the number of the corresponding states in the 10
transponder window for the transponder being sampled at the applicable
measurement update time.

C-8
 Table C-8
Definition of matrix HRR

Element State Definition

1,2 %Y H2

1,10 8H H3

1,7 8V x H4

1,8 sV. HS
1,9 8Vz H6
1,70 8EcA. -1
All other elements are zero.

[H1 H2 H31 =

OT-A)2'AX + (?T-A)( CT--VAYA + CkT--kA) (2 -- A)9Az - ^

('IT - YA)V'AY + (kT - iA) (XT - f(A) AX + (YT -'kA) (Zr - ZA)VAZ - AYRC

(- 2 ,A, + (Z- 20A) (T - "A)VAX + (Zr - ZA) (YT -YAAY - AZ

CN(L,& 0T(~)~ (C -2)

9T RT-A kT "A CE4

[H4 H5 H]= T- -- R_ - r R'AJ C (LX,) (C - 3)

Notes: 1. (XT, 'T, Zr) corresponds to the transponder position used in the range
measurement update.

2. H" transformation is required to transform from INS position and veloc-

ity frame to ECEF frame.

3. CE is defined in Eq (2-6).

C-9
 Appendix D
INS and Baro Altimeter CIRIS Flight Data

This appendix contains the CIRIS flight INS position, velocity, acceleration, and

wander azimuth angle (alpha) time histories; the baro altimeter altitude rate time history

is also included. This data is to drive the full ordered Kalman filter simulation. The data

is shown in Figures D-l thru D-11.

D-1
33.6

33.4

LATITUDE
33.2

33

32.8

32.6

32.4

32.2 I
24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D- 1. Latitude Trajectory for CIRIS Right

D-2
 -106

-107

-108

/ -109

S-110

LONGITUDE
-112

-113 F
24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D-2. Longitude Trajectory for CIRIS Right

D-3
 27000

24000

21000

18000

FEET
15000

12000

9000

ALTITUDE
6000

3000 I I I I I I
24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D-3. Altitude Trajectory for CIRIS Flight

D-4
 95

94.5

94

WANDER AZIMUTH ANGLE

93.5 (ALPHA)

W 93

12 92.5

92

91.5

91 FII
24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D-4. Alpha Angle Time History for CMRS Flight

D-5
 180
BARO ALTIMETER ALTITUDE RATE
150

120

90

60

30

FPS c)
-30

-60

-90

-120

-150 I
24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D-5. Baro Altimeter Altitude Rate Time History for CIRIS Flight

D-6
 600

X VELOCITY

400

200

FPS 0 -- F -_______

-200

-400

-600
24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D-6. X Velocity Time History for CIRIS Flight

D-7
 600

40O
4Y VELOCITY

200

FPS 0 --- __ A

-200

-400

-600
24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D-7. Y Velocity Time History for CIRIS Flight

D-8
 1 50

120L
Z VELOCITY
g0

60

30

FPS 0 1

-30

-60

-120

24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D-8. Z Velocity Time History for CIRIS Flight

D-9
 60

40 X ACCELERATION

20
C14

cj 0T

-20

-40

-60 II
24000 27000 30000 33000 35000 39000 42000 45000
SECONDS

Figure D-9. X Acceleration Time History for CIRIS Flight

D-10
 40

30 Y ACCELERATION

20

10

0
r1

-10

-20

-30

-40

24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D-10. Y Acceleration Time History for CIRIS Flight

D-1 I
 60

50Z ZACCELERATION

40

S30

20

10

0
24000 27000 30000 33000 36000 39000 42000 45000

SECONDS

Figure D- 11. Z Acceleration Time History for ChRIS Flight

D-12
 Bibliography

1. Air Force Avionics Laboratory. PROFGEN - A Computer Program for Generating

Flight Profiles, Technical Report AFAL-TR-76-247, November 1976.
AFWAL/AAAN-2, Wright-Patterson AFB, OH.

2. Breaux, Harold J. An Efficiency Study of Several Techniques for the Numerical

Integration of the Equations of Motion for Missiles and Shell, Report No. 1358,
Ballistic Research Laboratories, Aberdeen Prooving Ground, Maryland, February
1967.

3. Britting, Kenneth R. Inertial Navigation System Analysis, New York:Wiley-

Interscience, 1971.

4. Butler, R. R. And Rhue, G. T., "Kalman Filter Design for an Inertial Navigation
System Aided by Non-Synchronous Navigation Satellite Constellations," Masters
Thesis, Air Force Institute of Technology, Wright-Patterson AFB, Ohio, March, 1974.

5. Carlson, Neal A. and Stanton H. Musick. MSOFE User's Manual: Multimode

Simulation for Optimal Filter Evaluation, October 1987. Contract F33615-86-C-
1087, Subcontract 2013-SI, Avionics Lab, AFWAL/AAAN-2, Wright-Patterson
AFB, Ohio.

6. Cox, D. B. "Integration of GPS with Inertial Navigation Systems," Global Positioning

System, Papers Published in Navigation, Volume I, The Institute of Navigation,
Washington, D.C., 1980.

7. Cubic Corporation. Range/Range-Rate Subsystem (RRS) for the Completely Inte-

grated Reference Instrumentation System (CIRIS), AFSWC TR-7321, Cubic Cor-
poration, 9233 Balboa Avenue, San Diego, Ca., August 1973.

8. Defense Mapping Agency. Department of Defense World Geodetic System 1984.

Defense Mapping Agency Technical Report, No. DMA TR 8350.2. Washington,
D.C., 30 September 1987.

9. Dyrd, D. Litton Design Engineer, Telephone Conversations, Litton Guidance and

Control Systems, Woodland Hills, CA., Comm: 818-715-2991, April 1987.

10. Guidance and Control Division. LN-39 Standard Inertial Navigation System
ANLANi4I, Technical Description, Litton Systems INC., 5500 Canoga Avenue,
Woodland Hills, Ca., September 1987.

11. Guidance Test Division. "CIRIS II Test Summary Report," In-House Report, CIGTF,
6585th Test Group, Holloman AFB, NM, May 1988.

bi- 1
 Bi bliogra ph

12. Guidance Test Division. "The Completely Integrated Reference Instrumentation

System (CIRIS)," Unpublished Contractor Informational Pamphlet, CIGTF, 6585th
Test Group, Holloman AFB, NM.

13. Huddle, James R. "Inertial Navigation System Error Model Considerations in Kalman
Filter Applications," Advances in the Techniques and Technology of the Application
of Nonlinear Filters and Kalman Filters, AGARD-AG-256, Advisory Group for
Aerospace Research and Development, 13-1:13-19 (March 1982).

14. Intermetrics Incorporated. Computer Program Detailed Technical Description CEI-

1-72-B Post-Flight Processor for Ciris, Volume Ill, Part H1, Contract End Item
Specification, Intermetrics Incorporated, Cambridge, MA, 7 May 1974.

15. Lewantowicz, Lt Col Zdzislaw H. Deputy Department Head, Department of Elec-

trical and Computer Engineering, Personal Interviews, Air Force Institute of Tech-
nology, Wright-Patterson AFB, Oh, 28 March thru 21 April 1988.

16. Maybeck, Peter S. Stochastic Models. Estimation, and Control, Volume I, New York:
Academic Press, 1979.

17. Maybeck, Peter S. Stochastic Models. Estimation. and Control, Volume II, New
York: Academic Press, 1982.

18. Simkin, Gordon. CIRIS Post-Flight Performance Engineer, Personal Interviews,

Central Inertial Guidance Test Facility, Holloman AFB, NM, 1-3 March 1989.

19. Smith, Carl 0. Performance Characteristics of Integrated Inertial Navigation Systems,

Masters Thesis, Air Force Institute of Technology, Wright-Patterson AFB, Oh,
December, 1985.

20. Solomon, Joseph K. "CIRIS Special Study," Final Report, EENG 699, Air Force
Institute of Technology, Wright-Patterson AFB, OH, March 1989.

21. Speiser, M. "Systems Engineering Analysis Report for the F-16 Standard Inertial
Navigation Unit," Litton Guidance and Control Systems, Woodland Hills, CA.,
Revision B, March 1985.

22. Walker, Robert, Charles Gregory, and Sunil Shah. "MATRIXx: A Data Analysis,
System Identification, Control Design, and Simulation Package," Control System
YAgga&jj, Volume 2, Number 4, 30-36 (December 1982).

bi-2
 Captain Joseph K. Solomon
He served as an enlisted electronic technician in the Air Force from 1975 to 1981. At this
time he entered the Airman Education and Commissioning Program (AECP) and earned a
Bachelor of Science degree in Mechanical Engineering from the University of Colorado in

May 1984. Upon graduation he received a commission in the Air Force by attending the

USAF Officer Training School (OTS). His first assignment was as part of the engineering
team at the U.S. Air Force's High Speed Test Track in New Mexico from 1984 to 1987. In

June of 1987, he entered the Air Force Institute of Technology to pursue a masters degree
in electrical engineering, specializing in the aircraft navigation and stochastic control areas.
He is a member of the IEEE and ASME.

V-i