AD/A-002 879

DOPPLER VELOCITY SENSOR APN-200 ERROR

MODEL AND FLIGHT TEST RESULTS

William S. Widnall, et al

Intermetrics, Incorporated

Nat"l TWA*" wuld".Sob

U.S.01ARMININ OF CWMOCE
 SECUrOI TY CLA$,-.IF'Ir ATI,-I HorTHIS PAGE t'he-t Datn"nt•€l)

REPORT DOCUMENTATION PAGE RIRIADco,%irI.I.TING

nmreoRr.' '8INTiUC'lNG ORM
F'ORM%
. .LASCTR_7-9r
NU*L, 2. GQVT ACCESSION NO. 3. RECIPIENT'S CATALOG NUMBER

AUST -TR-7 4-9

REPORT & PERIOD COVERED

DOPPLER VELOCITY SENSOR APN-200 ERROR Contract

',Ot)EL AND FL11-11T TEST RESULTS Report • Technical
6. PERFORMING ORG. REPORT NUMBER

IR-78-74
7. AU9tON(s) B. CONTRACT oR GRANT NUMBER(s)

William S. Widnall and Peter A Grundy F29601-73-C-0089

1,. H jc 1 w,;hIfl
Oý,At3I2tATIQtl NAME ANW ADURISS 10. PROGRAM ELEMENT. PROJIC'I, TASK

AREA & WORK UNIT NUMUERS

intcrmctrics, Inc. 688G
701 Concord Avenue
II. C. R•LLI'c, Oi FILL :I. AV- AND AD ,'-"SS 12. REPORT DATE
ALr Force Spccial Weapons Center(PMRC) 8 March 1974
iK rt-land Air Force Base, Now Mexico 8 7 1 1 5 13. NUMBER OF PAGES

14 ?'C.',.TOttJ$ ,.'5 ,.r Y tfA*AL & AODItL:%%'i ,ttarent frem CrolgnA Ol lice) 15. SECURITY CLASS. (of this report)

6585th Test Group Unclassified

SAirForce Special Weagpons Center
1lololnan Air Force Base, New Mexico 8833 ISa.LDECLASZ1FICATION/DOWNGRADING

16. D)17t RIBUTI Ot ST.', tý.tXE T (of this Rf:.-0t)

Approved for public release; distribution unlimited.

•7. DISTM!I5UTIO- STATE!V',EŽ, T (of the abarr..ct entered In Balock 20, It different from Report)

Same as above

IS. SUPPLEMENTARY tKOTLS

19. t:F'Y WORDS (CliI-p#1:Ir on r.ev.,,se q•dp iI necesAiry eand identify by block uwlnber)

Doppler Velocity Sensor

Teledyne-Ryan APN-200 .
Doppler Error Model NATIONAL
NATION TECHNICAL
INFORMATION SERVICE
SERVICE
Doppler Flight Test Results us Sp•qt4•lf
.,,-,,w of--om.ce
. VA. 221SS

20. AFISTn/.CT fCcntinruo on reverse side it necessary and identify by block number)

The individual sources of error in a Teledyne-Ryan APN-200

Doppler velocity measurement are discussed. Also included is
the velociLy error due to INS attitude reference error. A
complete Doppler error model is constructed, including estimates
uf tiu CJ;L•:-cma val.ue.; of error parameters. Error model
sj.mnlificQ.ticn for 1zil'ian filter synthesis is discussed. Data,
from a 7 August 1973 flight test of an APN-200 at |iolloman Air
(ovor)
"•".'!~ 147: r,,u,,ioi ov
I ,VonsoLPTr ' Unclassiied

SICuRITY CLASSIFICATION OiF THti PACK (II404%OMA W,*4

 S CURITY CLA.StFICAyION OF THIS PAGEMIýh. Date Envfed)

Block 20

Force Base, are analyzed. The Doppler velocity performance,

including random noise level and slowly varying error, is found
to be in agreement with the accuracy predicted by the error
model.

}.U

YO~ O ~iiS
a~a~1
1~CO3YVCL AC~Wha . £

UnclaSriEfied
 FOREWORD

This technical report has been prepared by Intermetrics

for the 6585th Test Group, Holloman Air Force Base, New
Mexico, under Contract F29601-73-C-0089. The technical
monitor of this contract has been Capt. Walter G. Murch.

The discussions of Doppler sensor principles and

functional desiqn (Chapter 2), sources of Doppler velocity
error (Chapter 3), and the summary and conclusions (Chapter 5)
were written by Dr. William S. Widnall. Statistical analysis
of the Doppler flight test data was performed by Mr. Peter
A. Grundy, who also summarized the results (Chapter 4). The
manuscript has been prepared by Ruth Lepson and Eileen Martin.

iii
I ",,•{ L|LTRICS INCO•iPORATE.D - 701 CONCORD AVENUE •CA,,BRIOGE. MASSACHUSETTS 02138 -(017) 661-1840
X,.

ABSTRACT

The individual sources of error in a Teledyne-Ryan

APN-200 Doppler velocity measurement are discussed. Also
included is the velocity error due to INS attitude
reference error. A complete Doppler error model is
constructed, including estimates of the one-sigma values
of error parameters. Error model simplification for Kalman
filter synthesis is discussed. Data, from a 7 August 1973
flight test of an APN-200 at Holloman Air Force Base, are
analyzed. The Doppler velocity performance, including
random noise level and slowly varying error, is found to
be in agreement with the accuracy predicted by the error
model.

IV

NTER-AETRICS INCORPOHATED 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138. (617) 661-1840
•. "CONTENTS

4 Page
U 1. INTRODUCTION
2. DOPPLER SENSOR PRINCIPLES AND FUNCTIONAL DESIGN 3
W 2.1 Single Beam Doppler Shift 3
2.2 Measurement of Vector Velocity 4
2.3 APN-200 Functional Description 7
2.4 Performance Limits 9
3. DOPPLER ERROR MODEL 13
3.1 Introduction 13
3.2 Noise Due to Finite Beamwidth 14
3.2.1 Doppler Bandwidth and Correlation Time 14
3.2.2 Variance of Single-Beam Velocity Error 16
3.2.3 Three-Axis Covariance of Velocity 18
Errors
3.2.4 Significance of Doppler Noise Error 20
3.3 Tracker Time Constant 22
3.4 Beam Direction Errors 23
3.4.1 Residual Errors After Compensation 23
3.4.2 Temperature Effect 30
3.4.3 Installation Alignment 31
3.5 Terrain Bias 32
3.6 Surface Motion 38
3.7 Other Doppler Sources of Error 39
3.7.1 Transmitter Frequency 39
3.7.2 Receiver Cross Coupling 39
3.7.3 Frequency Tracker 40
3.7.4 Propagation Velocity 40
3.8 Non-Doppler Attitude Errors 41
3.8.1 Definitions 41
3.8.2 INS Stable Platform Attitude Error 43
3.8.3 INS Attitude Readout Error 44
3.8.4 INS-Doppler Installation Alignment 45
3.9 Combined Velocity Error Matrix 47
3.10 Doppler Error Model Summary 49
3.11 Error Model Simplifications for Kalman 51
Filter Synthesis

"V
" FI
01f./ ncs INCOBP-.MA I W
D 701 CO. iCOHD AVE NUE F CAM1.3110(Th, MASFACEI U.13-. I S 02-138 IF(617) G";1~1840
 page

4. ANALYSIS OF FLIGHT TEST DATA 55

4.1 Introduction 55
4.2 Isolation of Doppler Sensor Errors 57
4.3 Segment Selection for Statistical Analysis 65
4.4 Statistical Analysis 74
4.5 Flight Test Results Discussion 75
4.5.1 Doppler Noise 75
4.5.2 Slowly Varying Errors in Level Flight 83
4.5.3 Slowly Varying Errors in Maneuvering 84
Flight

5. SUMMARY AND CONCLUSIONS 87

REFERENCES R-1

vi

INTERMETRICS INCORPORATED • 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138 (617) 661-1"
 ILLUSTRATIONS

Figure
2-1 Four-Beam Doppler Ge )metry 5
2-2 Block Diagram of APN-200 Doppler Velocity 7

Sensor

3-1 Finite Doppler Beam Width 14

3-2 Terrain Bias Effect 33

3-3 Teledyne Ryan/NASA Reflectivity Data 34
3-4 Simultaneous Lobing 36

4-1 Aug 7 Flight Ground Track 56

4-2 Reference System Horizontal Velocity lo 58
Error (Computed by PFP Smoother)
4-3 Reference System Vertical Velocity la 59
Error (Computed by PFP Smoother)

4-4 Heading Velocity and Flight Events 66

4-5 Heading Velocity Error Coefficient 67

4-6 Drift Velocity Error Coefficient 68

4-7 Vertical Velocity Error Coefficient 69
4-8 Histogram of Heading Velocity Errors in 77
Level Flight
4-9 Histogram of Drift Velocity Errors in Level 78
Flight
4-10 Histogram of Ileading Velocity Errors in 81
Turning Flight
4-11 Histogram of Drift Velocity in Turning Flight 82

INTERMETRICS INCORPORATED- 701 CMNCOMD AVENUE • CAM._RIDG-. MASSAC.4UGF.TTS 02138 - (617) 661,4840
 TABLES

Table Page

3-1 Terrain Bias Velocity Error 37

3-2 Velocity Error Matrix Elements Due to all 48

Sources (Une-Sigma Values)

4-1 Sample Means and Standard Deviations for 76

Doppler Error Sources - Straight and Level
Segments
4-2 Correlations for Doppler Errors - Straight 79
and Level Scgments

4-3 Sample Means and Standard Deviations for 80

Doppler Error Sources - Turning Segments

&~ 0?0
7TRf~IC
r,ýConCC. AvjpiýL'E: GA9_Aj3OrCE. UkSSAr',s4UIEYTT 02111 f617.) 66-81
 CHAPTER 1

INTRODUCTION

The CIRIS (Completely Integrated Reference Instrumenta-

tion System) has been developed by floll'rran Air Force Base to
provide a highly accurate aircraft n-vigation reference for
the test and evaluation of advanced guidance and navigation
systems. The CIRIS flight hardw:are as originally implemented
includes a Litton CAIKS inertial navigator with barometric
altimnter, a Cubic Corp. CR-10o precision range/range-rate
measurement system, and a Hewlett Packard 2100 general purpose
digital cowputer. Intermetrics has provided a Post-Flight
Processor (PFP) -- a system oif computer programs -- for the
optimal estimation of aircraft position, velocity, and attitude
from the flight-recorded inertial and precision-ranging data.
The opc-rtio- oý the PFZ (the Kalman filter)
has been transferred to the onboard computer to provide in-
flight real-timue accurate indications of the aircraft naviga-
tion state.
Recently a Tzledyne-RyKn APN-200 Doppler velocity
sensor has been u2roct'red and has been added to Ute CRIS
flight har*&zare. Intfrmetrics has been asked to modify the
CIRIS Post-Flight Processor to incorporate the Doppler measure-
ments into the optimal etimation of aircraft navigation state.
A first step that is required is the selection or development
of a statistical error mldel rcpresentine, the sources of
Velocity err-or in the APW-200 Doppler velocity sensor. This
report sumxrtari:'s this first. st"p

Chapter 2 discusses qonoral Doppler sensor princip" s

plua the specitic functional &esiqnqPrf-200.
of the The single-
bo• Doppler shift and then the measure-*ent of vector velocity
using multiple bear.ms are discussed. A brief functional
description of the trc-200 provides nezessary background for
the discassion of sourcet, of error. Performrnce limits that
are im-plied by the functionaC design are discussed.

70i
t CAAMR2'7 'ASSAC I S87 IeIe-
1T5 3
 Chapter 3Wdiscusses the individual sources of error in
a Doppler velocity measurement. Many of the sources of error
are a function of the design of the Doppler velocity sensor.
In addition, several external environment factors are sources
of error. Also included in the error discussion is the
velocity error due to attitude reference error. The emphasis
is placed on identifying the physical parameters which cause
the velocity errors. With this physical understanding, one
can construct a Doppler error model which exhibits the correct
dependence on flight variables such as speed and attitude.
Also the number of error parameters to be determined from
flight test data can be held to a minimum.
Chapter 4 summarizes statistical analyses of the
velocity data from a flight test of the APN-200 out of
Holloman AFB, using the other CIRIS subsystems to estimate
the true aircraft velocity. The estimated errors of the CIRIS
reference velocity are presented. The flight is divided into
several segments for individual analysis. The measured
differences between the Doppler velocity and the CIRIS reference
velocity are analyzed with respect to velocity random error
and velocity slowly varying error. The measured statistics
are compared with the Doppler error model of Chapter 3.
Chapter 5 presents a summary and the conclusions.

UMTEAU.ZR.CS MCORIMPATED - 701 COCORID AVEXUE - C, idGO. UASSAEhUITTS 02138 - (617) 661.
 CHAPTER 2

DOPPLER SENSOR PIUNCIPLES AUND FUNCTIONAL DESIGN

2.1 Single Beam Doppler Shift

A Doppler velocity sensor may radiate one or more beams
of electromagnetic energy toward the surface of the earth.
Some of th-e energy illuminating the surface is scattered back
toward the Doppler sensor. If there is relative motion between
the Doppler antenna and scatterers on the surface, then the
energy returned is shifted in frequency. This is called the
Doppler shift. The Doppler frequency shift for a given beam
is

= 2 v cosa (2-1)

where A is the wavelength of the carrier-frequency radiation,

v is the magnitude of the aircraft velocity vector relative to
the surface, and a is the angle between the antenna beam and
the velocity vector. The frequtmy is shifted on transmission
and again on reception, which accounts for thc factor of two
in Equation 2-i.
The constant 2/A is called the Doppler sensitivity s
to the aircraft velocity compnonent along the beam a:is. In
togtas of the speed of licht c and the carrier frequency f0 ,
the sensitivity may also be expressed as

s 2/1 2f /c (2-2)

Yor the APN-200 with a transtnission frequency of 13.3 G11z,

the Doppler scns&tivity is 45.6 Hiz/knot.

-3-

'nur~ ~&s ~:Cow¶cmAD.

f CCiC3Ri)DWNi IJASACtWiIISC 102138- (61?) b661¶040
 2.2 Measurement of Vector Velocity
By measuring the Doppler shift along three or more non-
coplanar directions, one can calculate the three orthogonal
components of aircraft velocity. The APN-200 generates four
beams at fixed orientations with respect to the aircraft, as
illustrated in Figure 2-1, from Reference [1]. The figure
also shows an aircraft-referenced orthogonal coordinate system.
The subscripts h, d, and z denote the along-heading, cross-
heading (or drift), and down components respectively. Positive
directions are forward, right, and down. In terms of the three
orthogonal components of aircraft velocity, the Doppler shift
along a beam is

f = s(vh cosvh + vd cosvd + vz cosvz) (2-3)

where v I Vd' Vz are the angles between the coordinate axes and
beam ax s.
The beam orientations in the APN-200 have fore-and-aft
and lateral symmetry. Hence thc direction cosines for all
bears are the same, except for sign. The nominal Doppler
shifts are

fl = s(+X vh- m vd + n vz)

f2 = S(-k vh - m vd + n vz) (2-4)

3"f= s(-k vh + m vd + n vz)
f4 = s(+k 1h + m Vd + n vz)

where t, m, n are the magnitudes of the beam direction cosines.

It is clear that the desired aircraft-referenced velocity com-
porients cxn be computed as a function of the four Doppler
shifts as

T~(+ f 1
Vh-421hTs - f 2 -f3
_ + f 4)

Vd =s-j (- fl " f 2 + f 3 + f 4 ) (2-5)

• 1
- V -sn (+ fl++ + L3 + f 4 )

-4-

INTERMETRICS INCORPORATED • 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138 • (617) 661-1F
 It. -v
Zai

CY

S-~- A

I
3- H

Fig 2- /rBaiDplrGoer

/5
ITRETISINOPRAE -71COCR AEUE-CMBIG, ASAHSTT 238-(67 6114
 1b' -r'----- -

Alternate solutions for the velocity components are

+fl f) or (f3+ f4
Vh = (+f 1 f 2 3 4
Vd - + 3 o - + f4)
12-6)

V=f + f) or 2s (+ f 2 + f

The APN-200 mechanizes the four-beam solution (Equation 2-5)

when all four Doppler beams are usable. If one beam becomes
unusable, the system automatically switches to the appropriate
three-beam solution from Equation 2-6. When all four beams
seem usable, the redundant information provides a consistency
test, because in theory from Equation 2-4

1 f2 + f 3 - f4 = 0 (2-7)

For the angles defined in Figure 2-1, the direction

cosines are

2= siny = sine cos= cosA sint

m = cosy sini = sinO sinO = sinA (2-8)

n - cosy cosW = cose = cosA cos&

The nominal beam angles of the APN-200 are

y - 200 26' • 204 53

= 120 23t A 11 35' (2-9)
6 w 230 461 * 290 54

for which the nominal values of the direction cosines are

L = .349
m u, .201 (2-10)

n P.15

-6-
INTERMETRICS INCORPORATED 701 CONCORD AVENUE - CAMBRIDGE. MASSACHUSETTS 02138- (617) 661-IP
 2.3 APN-200 Functional Description

A block diagram of the APN-200 Doppler velocity sensor

is presented in Figure 2-2, from Reference [I].

Transmn~tt~r

Trt.

-. ' 'I t

Figure 2-2

Sensor[13
Block Diagram of APN-200 Doppler Velocity

"-7-
•)701 CONCORD AVEN UJ[ CAM QRIDGE. W•ASSACH IJ$,G:TS 02138 - (617) 661-164,3
*T[f•E1RIC S{['COROFHA][
 Doppler-frequency measurements are obtained sequentially
(not simultaneously) for each of the four beam directions.
There are two antenna arrays, one for transmitting and one for
receiving. The transmitter array has two ports. When port 1
is energized by the transmitter, energy is radiated along the
diagonally-opposite beams I and 3. When port 2 is energized
the transmitter energy is radiated along the other diagonal
beam-pair 2 and 4. The receiver array also has two ports.
The energy received from the tvo forward beam directions is
channeled to one port. The energy received from the two aft
beam direction is channeled to the other port. A solid state
switch is used to switch the transmitter between transmit-
array input ports 1 and 2. Similarly, a switch, at the Doppler-
frequency-tracker input, alternates between the outputs of the
receivers at port I and 2 of the receive array. The four
possible combinations of the two transmit-switch states and
the two receive-switch states produces the four separately-
selectable directions for Doppler measurements. The switches
are cycled such that 62.5 msec is devoted to the measurement
at each direction. The sequence of four measurements takes
0.25 sec. A consequence of this design is that three of the
Doppler shifts used to compute current velocity are old, intro-
ducing a small lag into the indicated current velocity.
The receiving antenna array is configured to form pairs
of lobes for each beam. Through the use of microwave hybrid
mixers, the phase sums and differences of the pairs of lobes
are provided for data processing. The sum of a pair of lobes is
a pencil beam with beamwidths somewhat greater than a single
lobe. The difference of a pair of lobes generates a uniquely
defined null beam axis. This null axis is determined solely by
array geometry; it is not influenced by terrain reflectivity.
By tracking the null, Teledyne Ryan indicates terrain bias is
substantially eliminated from the measured forward component
of velocityill.
Energy received from each radar beam is detected by
heterodyning it with a portion of tite transmitted signal
(zero-frequency super heterodyne detection) and is amplified.
A portion of the energy received from the sum radar beam is
detected as a reference Doppler signal in one mixer detector.
The remainder is detected ip phase quadrature with the refer-
ence signal in a second mixer.

INERMETRICS !tNVCOPORATED. 701 CONCORD AVENUE - CAMBRIDGE. MASSACHUSETTS 02138 - (617) 661-Ar
 -,- U •l •l -qk .. __ __

IZ

The Doppler spectra from the receivers are acquired and

tracked in the frequency trackc.. Only one frequency tracker
is utilized. It is sequentially switched to the appropriate
receiver. Teledyne Ryan states that this single channel
design saves space, weight, and power, and increases
reliability~l]. For each beam, the frequency tracker searches
the band of expected Doppler frequencies, acquires the Doppler
signal, and tracks the center-of-power of the spectrum. The
energy received from the difference arm of the microwave hybrid
is detected in a third mixer. The Doppler data from the
difference channel is cross-correlated with data from the sum
channel by the difference tracker to produce a signal fre-
quency in a digital frequency synthesizer corresponding to the
beam null Doppler frequency.
Each beam Doppler signal is independent and is present
in the tracker one-fourth of the time. At the end of each
beam time-shared period, information pertaining to that beam
is stored in tracker memory to be recalled when the respective
beam signal command appears. This makes acquisition independ-
ent for each signal.

The time-shared tracker produces output serial binary

statements of valozity along each of the four beams. The con-
verter performs serial binary arithmetic to compute the ortho-
gonal velocity components in aircraft coordinates. The con-
verter transmits the velocity data to the user (usually a
general purpose digital computer) in the form of word-serial
bit-serial binarl numbers.

Antenna boresight calibration data are determined at

the factory and are encoded in a read-only memory for trans-
mission to the user upon interrogation. The calibration data
are applied by the user to the uncorrected sensor velocity
outputs.

2.4 Performance Limits

The four beams are clustered near straight down so that

the aircraft may roll and pitch through substantial angles
without loss of any beam return signal. It is clear that no
return signal u-ill Le detected if a beam i3 above horizontal.
For the APN-200 angjles (Equation 2-9) the roll attitude and
pitch attitude that cause horizontal beams are roll ± 780
(at zero pitch) and pitch • 690 (at zero roll). The actual

-9-

rul"A."MICS INCOf4PORATC) - 701 CO:(. 011) AViN1 FE • CAMiIDG. 1AASACHI' 1SETTS 02138 • (617) 661 .iA40
 ...... 7

attitude limits are smaller than these absolute limits be-

cause the return signal becomes too weak for detection well
before a beam reaches horizontal. The attitude at which
signal is too weak is a function of transmitter power,
antenna gain, the terrain backscattering coefficient, the
angle of incidence, the altitude above the terrain, and the
receiver noise. The APN-200 is specified as functioning
through ± 450 in roll and ± 250 in pitch at 40,000 feet
altitude. At a lower altitude of 5,000 feet the limits in-
crease to ± 600 in roll and ± 300 in pitch. The above limits
are for a worst case surface, which is smooth water
(Beauford-one sea state). For rough water or land, larger
attitude excursions should be possible without loss of signal.
A consequence of clustering the beams near straight
down is the reduction in Doppler sensitivity to forward and
lateral velocities. The Doppler sensitivities for the APN-200
to forward, lateral, and vertical velocities are

Sh =s = 15.9 Hz/knot

s sm 9.2 Hz/knot (2-11)

s - sn = 41.8 Hz/knot

The Doppler frequency tracker in the APN-200 does not

have a zero frequency offset. Therefore the tracker is unable
to distinguish between positive Doppler frequency (positive
velocity along the beam) and negative Doppler frequency
(negative velocity along the beam). The APN-200 sensor is in-
tended for aircraft applications where the velocity vector is
generally in the forward direction. The mechanization there-
fore can assume that the forward beam Doppler shift frequencies
(fl and f 4 ) are positive and the aft beam Doppler shift
frequencies (f 2 and f 3 ) are negative. The appropriate sets of
mechanization equations (where all measured Doppler frequencies
are positive numbers) are obtained by reversing the signs of
the aft-beam frequencies f 2 and f 3 in Equations 2-5, 2-6, and
2-7.

-10-

INTERMETRICS INCORPORATED • 701 CONCORD AVENUE - CAMBRIDGE. MASSACHUSETTS 02138 • (617) 661 -1P
• • •"'•?•~ ~ ~~~. .
,;i••¸ - ..
•• • • -• ,:•;, ••• ,,•c -. . ;••., ;~• ,.,.••. , -';75.'S.,, ,W•.•,•-/4n <• 'S ••,, *<.-,'c.•
fl'..

This mechanization is not appropriate for a helicopter,

which can fly backwards, sideways, or straight up and down.
In each of these cases, assuming the tracker finds the Doppler
signals, this mechanization wou.L•; qive an incorrect velocity
indication because either a forward beam is receiving a nega-
ti%- Doppler shift or an aft beam is receiving a positive
Doppler shift.
The frequency tracker has a high-pass filter at the input
to reject noise below 500 Hz. As a result, the tracker will
not find Doppler frequencies below about 500 Hz. The Doppler
frequency along each beam is of the form

f= sh vh sd vd sz vz (2-12)

For the APN-200 sensitivities given in Equation 2-11 and

because the Doppler frequency must be above 500 Hz, the vehicle
velocity components must satisfy

15.9 vh t 9.2 vd ± 41.8 v. > 500 Hz (2-13)

where the velocity components are expressed in knots. At zero

drift and vertical velocity, a minimum speed of about 30 knots
is required for the Doppler to function. Thus no Doppler
data are available at rest or at slow taxi speeds.
Drift velocity limits also can be inferred from
Equation 2-13. The worst case is at slow speeds and high
angles of attack. For example, if an aircraft has heading
velocity of 100 knots and vertical velocity of 13 knots, then
drift velocity cannot exceed 60 knots. This is a drift angle
limit of about 30*.

0,.:;;:<•;tO•Al'
H ) 701 CO ;CG1I)AV,',hJLt C\'.'[;RIDfl[r. I.'ASSACfIUS[VTI]S 02136. (617) 661.18410
 CHAPTER 3

DOPPLER ERROR MODEL

3.1 Introduction
This chapter discusses the various sources of velocity
error in a Doppler velocity measurement. Of interest are not
only those sources of error inherent in the electronics and
functional design of the Doppler velocity sensor, but also the
sources of error in the external environment. Furthermore, to
be useful for navigation, the aircraft-referenced measurement
velocity components must be transformed to navigation-frame
coordinates (such as an east-north-up set). Thus in addition
to the Doppler equipment and environment errors, there are
velocity errors due to attitude-reference errors.

If optimal filtering is to be used to mix

velocity-measurement data with inertial or other the Doppler-
navigation
data, then for filter synthesis the designer requires estimates
of the statistics of the Doppler-velocity-measurement errors.
Statistical parameters of interest include: RMS level of the
uncorrelated random error (noise) in each velocity component,
cross-correlation between component errors, uncompensated bias
magnitudes, and slowly-varying error RMS levels and correlation
times.
The emphasis in this chapter is placed on identifying
the physical parameters which cause the velocity errors. with
this physical understanding, one can construct an error model
which exhibits the correct dependence on flight variables,
such as speed and attitude. Also the number of error para-
meters to be determined from flight test data can be held to
a minimum.
Typical values for the sources of error
including estimates of values for the APN-200 byare mentioned,
Teledyne Ryan.
Measured parameter values from flight test data are presented
in Chapter 4.

-13-
"......... .(C)t OA1(0f) • -701 CONCCPQ AVENUE • CA MBI3;
DG E.MAqSACHUR!ClTS 02138 * (617) 6f1-1840
-- _- -- _" •.

3.2 Noise Due to Finite Beamwidth

3.2.1 Doppler Bandwidth and Correlat Jon Time

The indicated velocity from a Doppler velncicy sensor
is very noisy. This is related to the firtits width of each
Doppler beam, one of which is shown in Figure 3-1. The
single Doppler frequency of an infinitely narrow beam was
given in Equation 2-1. But with a finite beamwidth, the
Doppler spectrum is broadened. Energy returned by scatterers
at the forward edge of the beam will have a higher Doppler
frequency than that of the energy returned by scatterers at
the aft edge of the beam. The bandwidth Af of the Doppler
signal is given approximately by the differential (with re-
spect to look-angle m) of Equation 2-1

f s v Cosa (2-1)
Af = s v AQ sina (3.

where s is the Doppler sensitivity 2/X and Am is the beamwidth

(in the plane of the velocity vector and the nominal beam
direction).
An alternate expression which shows the dependence of
Doppler bandwidth upon the shape of the beam is [3]

Figure 3-1

Finite Doppler Beam Width

INTEI:LWETRIC.S INCORPORATED • 701 AVENUEi 4 -• CABRG.

"CONCORD MASCHSTTS 02138 -(617) 661- 1r
C1
S2 B2 2 n2 1 32
Af s[v, + v2 + -i--v 2

(3-2)
4 4)1/2
(8 + )l/

where

Af Doppler signal 3db bandwidth

s Doppler sensitivity 2/X

V1 , v 2 , v 3 velocity components in beam coordinates

mi,nor
= and major two-wav 3db radian
beamwidths

For the APN-200, the two-way minor-axis beaimwidth (fore-and-aft

direction) is 1.B0 (.032 radian) and the major-axis beauwidth
(lateral direction) is 3.40 (.06 radian)¶iJ. For small to
moderatc drift anr.les, the largest term in Eqluation 3-2 would
be the torward-velocity-minor-axis term.

The relative (fractional) spectrum width is frequently

of intercst, and is found by dividing Equation 3-1 by Equation 2-1.

Af = a tan a (3-3)
f

For a .032 radian boautwidth and a look angle of 70Q (APN-200

with zero drift and vertical velocity) the relative spectrum
widtl, is 9 percent. Other Doppler radars having wider beams
have relative spectrum -Jidt*hs of 15 to 25 percent[2).

The period of time during which the frequency and phase

of the Signal are approxirvatcly constant is called the Doppler
correlation tire. 'he correlation tir-. -, s pAroportional to
the reciprocal of the spectrum half width &l/2(4)

Tf 2/4w 1f(M1f) (3-4)

-15-

IV'....fl~C
:OW.r"NffiAT .... 7'01 .9N-CC V/WF, C • CAt,%WI•GL. LAA.,SACH4•ETTS 021Mt (38 , tl-,, :,
 Note from Equation 3-1 that the Doppler bandwidth is directly
proportional to vehicle velocity, and therefore the Doppler
correlation time is inversely proportional to velocity. For
the APN-200 with .032 radian beamwidth, look angle of 700,
sensitivity of 45.6 Hz/knot, and assuming an aircraft speed of
400 knots, the Doppler bandwidth from Equation 3-1 is 550 Hz.
The Doppler correlation time from Equation 3-4 is .0006 sec.
Fried[21 points out the Interesting fact that the Doppler
correlation time is of the order of the time required for the
vehicle to travel a distance equal to the length of the
antenna.

3.2.2 Variance of Single-Beam Velocity" Error

if the Doppler frequency tracker could follow the rapid
fluctuations of the instantaneous frequency random process,
then the measured velocity along the beam would exhibit very
iarge velocity random error. The one-sigma deviation of the
ins -ntaneous frequency random process from its mean value is
approxiýwately half the Doppler sigual bandwidth (41.

C.f af/2 (3-5)

For the 400 knot example, for which the Doppler bandwidth is
550 Hz, the one-sigma frequency deviation is 225 Hz. For the
Doppler sensitivity of 45.6 WA/knot, this is a one-sigma beam
velocity error of 6 knots.
To reduce the anplitude of the random velocity error it
is desire-able to filter the noise. A practical frequency
tracker has an output time constant tv that is greater than
the correlation time tf of the Doppler signal. The APN-200 has
output correlation of about .2 sec. Therefore the frequency
aauremant error out of the tracker has variance xed=_ed to

2 2
f f OfTV

-16-

INTEr--TfUCS ITCORPQRATE0 - 701 COWAM AVE-K-E - CAMBROGE. MASSACKJSETTS 02138 - J617) 661-,
Measured velocity is related to measured Doppler frequency
"by the Doppler sensitivity factor s. The error in measured
velocity is similarly related to the error in measured
frequency

av =Gf/s (3-7)

The velccity error of the APN-200 will be somewhat

noisier than implied b; the above equations. This is
because syste-ms employing lobing techniques (for terrain bias
error suppression) have random error variance increased by
a factor[2J

Af/Wf
A (3-8)

where 6f is the Doppler signal bandwidth and Wf is the lobing

low-pass tracker-filter bandwidth. This filer bandwidth is
an intermediate bandwidth, not the same as the narrow band-
width azsociaLed with the output correlation tine ' v. The
tracker bandw idth of the APN-200 is adjusted in steps as a
function of the teasured Doppler frequency. At 400 knots,
the single beam Doppler frequency is about 6400 Hiz. This is
in the rtiddle of the 4000 to OOOD 1iz range for which the
tracker bandvidth is set at 60 Ha. The Doppler bandwidth at
400 knots wras noted to be 550 liz, so the variance increase
factor (Equation 3-8) is about 10.

Combining the above relationships, the variance of

single-beam velocity measurement error is

rbb a 2 v 4a At sinn
VS v (3-9)

whore

v = -magnitude of vehicle velocity

S = mechanization variance-increase factor

boa-
ea•idth

a = angle between velocity vector and beam

s - Doppler sensitivity
TV output time constant

-17-
Ct- f- (2

• m•in i ml. l.'.lUlmm .11 *iIlllm AiDl

I..rS• I n 38lnmnmllW1•l9I IInI
 Note the velocity error variance is proportional to velocity
(assuming that the step changes in tracker bandwidth keep 0
(Equation 8) approximately constant). For 8 = 10, Aa = .032
radian, a = 700, s = 45.6 Hz/knot, and TV = .2 sec, the
APN-200 should have single-beam velocity error variance of
about .0026 knots 2 /knot. At 400 knots this would be 1.0
knots 2 or (1.0 knot) 2 .
The variance of the random error could be reduced by
lengthening the output time constant. However, this would in-
crease the lag error, which occurs when velocity along a beam
is changing.

3.2.3 Three-I-xis Covariance of Velocity Errors

The velocity random error of each beam is uncorrelated
with the errors of the other beams, because an independent
set of sca terers is providing the Doppler return for each
beam. Assuming small drift and vertical velocity, the four-
beam iýr:ay symmetry implies equal velocity-error variance for
each becuu.

The velocity errors 6 vh, 6vd, 6v, in the aircraft-

referenced velocity components, for the four-beam mechanization
of Equation 2-5, are

Svh (+v
1

6 vd= . (-.Sv 1 6-,v +6v 3 +6v 4 ) (3-10)

vz 'V 1 +6v 2 +6v 3 +6v 4 )

where dv 1 , 6v 2 , dv3p 6v 4 are the velocity errors of the four

beams, and L, m, n are the beam direction cosines. If the in.-
dividual beam errors have zero mean, then the cotm:onent errors
also have zero mean. It can be shown, using the assumed zero
correlation between beam errors, that the variances rh, rd, rz
of the corponent velocity errors are

-18-
I TE~d,JETRd6S tNCTiOi -E'01 CONCO@,0 AVENUE -A BRDE.• M,.SSA•JIUS..1'TS 02138.- (6"1,7 611 r
 2
rh = rb/( 4 £2

r = rb/(4m
2

rz rb/(4n2

where rb is the single-beam velocity error variance (Equation 3-9).

For the APN-200 direction cosines of .349, .201, .915 the com-
ponent variances are

rh =2.1 r
h b

rd = 6 . 2 rb (3-12)

r 0.3 rb

One might expect the cross-correlation between component

velocity errors to be non-zero, because the error of each beam
contributes to the error of all three components. However it
can be shown (by computing the expectation of the product of
pairs of equations in set 3-10) that the cross-correlation is
zero. This result is due to the four-beam symmetry whereby
positive correlation due to one beam error is offset by equal
but negative correlation due to another beam error. Thus the
covariance matrix of the aircraft-coordinate velocity vector
noise error is theoretically

th Rf
R b 0 6.2 .006a (3-13)

If the siagle-beri variance is about .0026 knot2/knot, as

discussed earlier, the covariance should be about

C X-19-

'... !,,. S.F

(:.•..3 AV,,.ED •7.1 :;C.:CO:•r),,,I.,U5: CAMB•RIEG:E, MASSACHUSETTS 02133 (617) 6G1 -1340

•. ...
 R v 0 .0161 (3-14)
0 .0 .0008

where v is in knots to obtain velocity error variance R in

knots 2.
The above variance values, obtained from theoretical
considerations, may be compared with the noise specifications
for the APN-200. The noise specifications are that the output
noise spectral densities Ph, Pd, Pz shall be less than .001,
.002, .0005 knot 2 i(radian/sec)/knot[l]. These are one-sided
power spectral densities in angular frequency that are related
to velocity error variance by[2]

r = vP r/Tv (3-15)

For the output time constant or correlation time TV of .2 sec,

the sDecified error covariance matrix is

[016 00
R = V .031 0 (3-16)
0 .008

The estimated component variances (Equation 3-14) are seen to

be smaller than the specification (Equation 3-16).

3.2.4 Significance of Doppler Noise Error

It is interesting to note that the position error due
to the velocity noise is a function of distance traveled but
is not a function of speed. Consider a Doppler navigator
that computes position by integrating velocity in a digital
computer. Assuming the computer samples velocity every At
sec, the position error due to velocity error at the end of n
sample intervals is

-20-

INTERMETRICS INCORPORATED • 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138 - (617) 661-1P
 n

6p E 6v. At (3-17)

Assuming uncorrelated random velocity error (At greater than

the error correlation time), the variance of the position
error is

2
rr6p = nr 6v At (3-18)

But the velocity error variance is proportional to velocity,

as indicated in Equation 3-14. Hence the position error
variance is of the form

r n v k At2 (3-19)

The distance traveled is

d = n v At (3-20)

So, Equation 3-19 also may be written

r d k At (3-21)

which shows that the position error variance due to the

velocity noise is not a function of speed.

The noisy nature of the output of a Doppler velocity

sensor .may be bothersome in applications where the instantaneous
indicated velocity must be very accurate. However in Doppler
navigation, where position error is of greatest interest, the
effect of the random noise velocity error is reduced to a
negligible level compared with other sources of error after
only a few miles distance traveled. The fractional one-sigma
position error due to velocity noise is from Equation 3-21

ar /d = (k At/d) 1/2 (3-22)

-21-
' ?LMETRCS ItNCOMPORAI ED 701 CONCORD AVENUG • CAMBRIDGE-, MASSAChiUSETTS 02138 • (617) 661-1840

k~
 Assuming a velocity error variance factor k at the specified
level of .016 knots 2 /knot, sample interval At of 1.0 sec
(1/3600 hr), and distance traveled d of 10 naut. mi, then the
fractional one-sigma position error is .0007 of distance
traveled. This is smaller than the error due to a typical
bias velocity error, which is of the order .001 of distance
traveled.

3.3 Tracker Time Constant

The Doppler frequency tracker in the APN-200 has a time
constant of about 0.2 sec. By having a noticeable time con-
stant, the tracker filters the very large velocity-error noise
inherent in the Doppler signal. However by having a noticeable
time constant, a lag ezror is introduced whenever velocity
along a beam changes.
In addition to the lag due to the frequency tracker
time constant, there is in the APN-200 a lag due to the single-
channel four-beam sampling mechanization. The effective time
constant due to both the tracker time constant and the sample-
and-hold mechanization is about 0.25 sec.
The lag error can be noticeable during speed changes.
Consider that the aircraft accelerates from 300 to 420 knots in
60 sec (about 0.2g acceleration). The tracker time constant
of 0.25 sec will cause a 0.5 knot velocity error, which will
persist during the 60 sec period.
In the absence of wind, the aircraft can maneuver at
constant speed without incurring a Doppler lag error. This is
because an aircraft normally flies at small angle-of-attack and
sideslip angle whether straight and level, or in a coordinated
turn, or climbing, or descending. The ground speed vector has
nearly constant ccmponents in aircraft-fixed directions and in
particular in the beam directions.
However it there is a strong wind aloft, the ground
speed vector may not be aligned with the longitudinal axis of
the aircraft. In this case roll rate or heading rate causes a
velocity lag error. Consider a cross wind of 100 knots and a
roll rate of 10 deg/sec (1/6 rad/sec). The tracker time con-
stant of 0.25 sec will cause a 4 knot velocity error for the

-22-

INTERMETRICS INCORPORATED . 701 CONCORD AVENUE " CAMBRIDGE. MASSACHUSETTS 02138 . (617) 661-11
few seconds that the aircraft is rolling at that rate.
Initially, at zero roll angle the velocity error is sub-
stantially in the vz "vertical" component, At steep roll
angles, the error due to roll rate is also seen in the vd
"drift" component.

In a shallow turn at a heading rate of 3 deg/sec

(1/20 rad/sec), the 100 knot wind and 0.25 sec tracker time
constant cause a velocity error of 1.2 knots. When the
track is across the wind, the error is mostly in the heading
direction. When the track is along the wind, the error is
mostly in the "drift" direction. The turn duration at
3 deg/sec for a 180 deg turn is 60 sec.

3.4 Beam Direction Errors

The center of each beam will never lie precisely along
the intended nominal direction, because of three factors:
1) normal manufacturing tolerances produce an antenna having
boresights slightly off of the nominal directions; 2) changes
in antenna temperature change the physical size of the
antenna and therefore shift the directions of the beams;
3) the antenna will not be mounted on the aircraft in precisely
the intended nominal orientation. These sources of beam
direction error and their effect on velocity error are dis-
cussed in this section. The terrain bias effect, which might
also be considered a beam direction error, is discussed in
Section 3.5.

3.4.1 Residual Errors After Compensation

The Doppler frequency shifts along each of the four
nominal beam directions were given in Equation 2-4. In terms
of the actual beam directions, the Doppler shifts are

f1 ki -ml n1 Vh

f2 -£2 -m2 n2 "d

-= S (3-23)
3 3 m3 3 z

4 4 4 4

-23-
S,.ji;,
S,,iC(.,m-OMA) • 701 CONCORD AV'tN.L • CAMBRIDGE. MASSACHUSETTS 02138- (617) 661-1840
 where the kj, m-~ n- are the magnitudes of the i-th beam
direction cosines. The velocity computation (Equation 2-5)
expressed in matrix form is

AVhi 1/ -l/L -l/t l1/

1/i 1

-I/m -1/m 1/m 1/m f2 (3-24)

z 1
1/n 1/n 1/n 1/n f3

f
LJ
4

where v denotes the computed velocity. Note the velocity

computation is based on the nominal direction cosine magnitudes
Z, m, n. Combining Equation 3-24 and Equation 3-23, the com-
puted velocity is related to the true velocity as

U 1+ Z2 +z 3 +Z4 )/4Z (-m+M 2 -m3 M4 )/42 ( nl-n2-n 3+n 4 )/ 4 1

(h vh
VdI= ( -£I+£2 -£3 + 4 )/4m ( m1 +m2 +m3 +m )4 m 1'-2 +n3 +n4 )/4 m ~
A

vz U£1-z2-k3+t4)/4n (-ml-m2+m3+m4)/4n ( nl+n2 +n 3 +n 4 )/4n V.

(3-25)

Note that if all actual direction cosines had the nominal

values, the above matrix would be the Identity matrix and the
computed velocity would equal the true velocity. If one re-
places the actual direction cosines in Equation 3-25 by the
equivalent nominal value plus error (i1 5ý k + 6tI etc.), it
can be shown that the computed velocity v and the velocity
error 6v are

iNICth-,THtiCS INCORPORATED 701 CONCORD A'/NUE CAMiBRIDGE. MASSACHUUL ... 1.2i33 k617) 6G1-1P
* I,.: ,I 1 I I lli~ l _ _ZZI - -•-"-Z•, •• • _

V (I + E) v (3--26)

6v = E v (3-27)

where E is the error matrix whose nine elements are

h= ( 61 + R2 + 6H3 + 62 4 )/4Z

Ehd = (-6mI + 6m2 - 6m3 + 6m4)/4Z

Ehz = ( 6nI - 6n 2 - 6n 3 + 6n4 )/4Z

Edh = (-6Z1 + o£62 - + z3)/4m

Edd ( 6m1 + 6Rm2 + 6m3 .+ 6m4)/4m (3-28)

Edz = (-6n - 6n 2 + 6n3 + 6n4)/4m

Ezh = ( 6£1 - H2 - 6£3 + 6£z4 )/4n

Ezd (-6m1 - 6m2 + 6m3 + 6m4 )/4n

E 6nI + 6n 2 + 6n + 6n 4 )/4n

The subscript convention in Equation 3-28 is Ehd is in the

h row and d column, such that 6 vh Ehd vd, etc.

-25-

tA 'D 7,W IN 2'.

vDrW
1 AG C;1 :." i.;I
f.'; 'SI:. - (,517) .'I-R,

k .l
 Assume that the one-sigma levels of the forward and
lateral boresight errors 6U and 6m are about 2 milliradian

= a = 0.002 radian (3-29)

m

The change in the third component 6n is not independent

because the sum of the squares of the three direction
cosines must be unity. It can be shown for small errors

6n = (-I 6k -mcm)/n (3-30)

Assuming 6k and 6m are statistically independent, the one

sigma value of 6n is
22 ~ 1/2
an 2 1 / /n (3-31)

which for nominal beam direction cosines 2, m, n of .349,

.201, .915 and a, am = 0.002 radian is a one-sigma error
;f

an = 0.00088 radian (3-32)

Assuming that the errors of any beam are statistically

independent of the errors of each other beam, the one-sigma
value of each element in the error matrix E can be shown to be

0/2.12 Oin/2Z an/2Z

a2 /2m a /2m an/2m (3-33)

aza/2n m/2n an /2n

-26-

ir Iti,4EMRICS I:NCORPORATED •701 CONCORD AVENUE - CAMBRIDGE, MASSACHUSL I o u, 138 (617) 661-IP
 For the assumed 2 milliradian a and am, these uncompr'nsated
one-sigma values are
k ,
.0029 .0029 .0013

= .0050 .0050 .0022 (3-34)

l0011 .0011 .0005

vL

The level of these boresight errors is significant,

so it is necessary to measure these boresight errors and com-
pensate the measured velocity accordingly. The antenna
boresight errors are measured at the factory by Teledyne Ryan
as part of the acceptance tests of each unit. This establishes
the values of the elements of the error matrix E. The compen-
sation equation is related to the inverse of Equation 3-26,
which is approximately
A
V= (I - E) v .(3-35)

(Note I-E is approximately the inverse of I+E in that

(I+E) ()-E) = I-E 2 , which is the identity matrix to first order.)
The one-sigma levels of the compensation matrix elements are
the same as given in array 3-34, since the compensation matrix
is simply the negative of the error matrix.
The actual compensation equation recommended by Teledyne-
Ryan for the APN-200 in use at Holloman AFB is

AA

vh 1+.00271 0 0 vh

Ag

vd = -. 00078 1-.00658 0 Vd (3-36)

A A

v -. 00146 0 1+.00016 V

-27-
•~5 ~~~
',.F.: , ,,, [ ,,icc);-COM•) AVE/,bi - '•.. ,- f.,..GI.L.!..,.,C
7,C
17 G6; -1840
 Note the levels of the compensation numbers are consistent
with the one-sigma levels based on the 2 milliradian bore-
sight-error assumption.
The recommended compensation equation matrix is non-zero
Only on the main diagonal and the first column. The assumption
behind neglecting the other elements is that vd and vz are small
compared to vh. However in a high cross wind v4 is not
negligible and at a high bank angle in wind vZ is not negligible
either.
To analyze the errors of the partial compensation
method, denote the error matrix E as being composed of
elements to be compensated EC and elements not to be compen-
sated Enc

E =EC + EOc (3-37)

The indicated velocity (Equation 3-26) is

= (I+Ec+Enc) (3-38)

The compensation (Equation 3-36) is

(I•E c) v (3-39)

where E contains the estimates of the error elements to

be compensated. Combining Equation 3-39 and Equation 3-38
and neglecting second order terms, the compensated velocity is
^, i+ ~ ^_c+nc
v = (I+E +E )Ev (3-40)

The compensated velocity error matrix E has some elements

from the original uncompensated error matrix and some elements
which are residual errors after compensation

E = Ec-c+Enc (3-41)

-28CA
iNi Li'M.t•I )RIGS INC)R.PORATEL).701 CONCORD afEtiUL ~•-FD5.MS•~t•L,•-,:18•(1)•-
The one-sigma value of each element in the compensated error
matrix El can be shown to be
0/212 am /2X an/21

£ a '/ 2m a '/2m an/ 2 m (3-42)

u,'/2n am/2n
m ni 2n
/n

where the un-primed siqw.as are the uncompensated beam directi.'n-

cosine errors and the primed sigmas are the one-sigma direction-
cosine errors after compensation. Estimates of the one-sigma
errors after conpensation are[l)

a'= 0.25 arc min .00007 radian

(3-43)
am 1.0 arc min = .00029 radian

Using ;quation 3-31, the third direction-cosine uncertainty

after co.mtpensation is

n radian S.00007 (3-44)

The one-sigma values of the elements in the compensated error

matrix E' (Equation 3-12) are

.00010 .0029 .0013

.00017
000 .- CI072 .0022 (3-45)
.00004 .0011 .000K4

-29-
S....
; I " .. •CC
. •• , .•,• i :- •." ... ; • •..;.'•°. #-I . •
*4%,I
 Note that the elements that were not compensated are
not necessarily negligible. A large source of error is one
that contributes a velocity error that is 0.001 of speed.
The column 2 row 1 error is at that level when vd = %h/3 ,
such as when flying at 300 knots in a 100 knot cross wind.
Only the five c.ilibration numbers shown in Equation 3-29
are encoded in the read only memory of the APN-200. However
the other four calibration numbers could be obtained from
Teledyne-Ryan. A complete calibration matrix with nine ele-
ments would require a negligible increase in computation for
the processor using the Doppler data.

3.4.2 Temperature Effect

t
Thermal expansion of the antenna causes a symmetrical
distortion of the beam positions. The temperature coefficient
of the direction cosines of the beams is 2.34 x 10-5 per
degree C. [1] The radome insulates the antenna from the very
lamre changes in atmosphere temperature from sea-level to
altitude. Air from within the aircraft is used to cool the
system. A typical deviation of the antenna temperature from
the temperature at which the antenna was calibrated would be
about 20 0 C. This would produce a shift in the direction cosines
of 0.00047 radian.
The velocity error caused by these direction-cosine
errors is also governed by Equation 3-27, whose error matrix
E is given in Equation 3-28. Assuming symmetrical distortion
of the beam positions (61's equal, 6m's equal, 6n's equal),
the error matrix is simply

0 0
0 6m/M 0 (3-46)
0 0 an/n

Assuming 6m equals 6U, 6n computed from Equation 3-30 must be

6n -0.6 6L (3-47)

-30-

Ni4-Wh. TRICS INCOWRPQATED- 701 CONCORD AVENUE - CAk•U3DGE. UASSAC#ULAL VI,138 - (617) 66' ir
 Then for the assumed one-sigma 20OC temperature deviation,
the one-sigma values of the elements of the error matrix are

.0013 0 0

E 0 .0023 0 (3-48)

0 0 .0003

This is one of the largest sources of error in the

APN-200. For applications where one wishes to reduce this
source of error, Teledyne-Ryan recommends mounting a tempera-
ture sensor on the antenna and using the temperature data to
"compensate the indicated velocity.

3.4.3 Installation Aliqnmenvt

Installation of the unit in the aircraft is referenced

to the mounting flange on the radar housing for pitch and roll
alignment and to yaw alignment pins in the aircraft which
mate with yat: aliqr~ment slots in the radar housinq. The
matinc! surface of the radar mourting flange has a tolerance
of 0.020 inches, which for the 25-inch long surface is equiva-
lent to pitch and roll installation tolerances of 3 arc min.tIf.
The yaw alignment pins fit with a tolerance of 0.005 inches
in azimuth, which is equivalent to a yaw installation tolerance
of 45 aic sec[l]. Assuming a rectangular distribution for the
error, between the negative and positive tolerance values, the
one-sigma value for the error is 0.58 of the tolerance. In
which case, the one-sigma alignment errors are

h 1.74 arc min .00051 rad

ad = 1.74 arc min - .00051 rad (3-49)

UZ = .43 arc min - .00013 rad

-31-

- .- , t , .• .' ., . i .% I A , I" .. • • ) : ., • &

 The velocity error caused by alignment errors ch, Cd, CZ
can be written in the matrix form (Equation 3-27) where the
error matrix in this case is

0 Cz -d

E =-z 00 0 •h(3-50)

LCd -Ch 0_

From Equation 3-49, the one-sigma values of the elements in

the error matrix are

0 .00013 .00051j

.00013 0 .00051 (3-51)

.00051 .00051 0

3.5 Terrain Bias

The scattering coefficient of the illuminated terrain

is a function of the angle of incidence of the radiation. The
radiation att tue near edge of the beam footprint is steeper
than the radiation at the far edge of the beam footprint. The
higher scattering coefficient of the near radiation causes a
shift in the beam center of power toward the vehicle, as shown
in Figure 3-2. For a Doppler sensor that simply tracks the
center of power, the shift in direction of center of power
causes a velocity error. This effect is called terrain bias.

For typical geometry Doppler sensors in level flight,

the error in heading velocity is of the ozdor[2)

6vh .001 ( )2 m v (3-52)

-32-
i,k,;. .... Mr,,
•,CC-APORTED 7,51 CO-NCOAD ..... -"CA DUG .IA,S•A,,•--. 33 -6k ")63 1.
where Aa is the beamwidth in degrees, m is the slope of the
reflectivity curve at the nominal incidence angle in decibels
per degree, and v is vehicle velocity.

CENTER OF
TRANrsITTED
• CENTER OF .
~~REFLECTEO POWER
POWER
HIGHER • LOWER

REFLECTIVITY \ PEFI'ECTIVITY

Figure 3-2

Terrain Bias Effect

The reflectivity slope varies greatly between various

kinds of terrain, as shown in Figure 3-3 fro.- Reference l].
Smooth water has the steepest slope. Rough or wooded terrain
has the smallest slope. In level flight the beam incidence
angle of the APN-200 is 24 degrees. At this angle some of
the reflectivity slopes are: .88 db/dec for smooth water,
.34 db/deq for smooth barren terrain, .06 db/deg for rough
or wooded terrain. For beamwidths of several degrees and for
the ovor-water slope, Equation 3-52 indicates that terrain
bias can be a very significant source of error.

-33-

,." . ON(O - AVfi • ,A ' , A(,- "....

- • '617)6,1-1840
;' -' 38138
 LEGEND
Significant Wave Assoclated

Height (Sea) Beaufort No. Code

13 Ft. B4 0
5 Ft. B3
10 -- 0. 5Ft. B1

Terrain (Land)
: ".=• Extremely Rough, Wooded

or Heavy Vegetation
0- Smooth, Barren Terrain

~•--.Smoothest Terrain Encountered

(Man-smoothed Areas, etc.)

-10

0-
m 15..
.

-20
C,,

-- S-25 , .,,.

-30
e
, GPL Water Surface 10
Condition 1 (8.8 GHz)

NRL Beaufort 1
(9.4 GH2)
-40 13
-30Giz - Ryan/NASA Report 57667-2, Sept. 1967
9.4 GHz - NIRL (Grant & Yaplee) Proc. IRE, July 1957
8.8 GHz - GPL "Doppler Radar Principles', McMahon,
- [ I
.I ,I_. I I I.I I July 1967
_ I, I ._______
0 10 20 30 40 so so 70

INCIDENCE ANGLE, DEGREES

Fig. 3-3 Teledyne Ryan/NASA Reflectivity Data[ 1 ]

-34-
INTERMETRICS INCORPORATED - 701 CONCORD AVENUE
• CAMBRIDGE, MASSACHUSETTS 02138. (617) 661-181
 To suppress the terrain bias effect in the heading
component, the APN--200 employs simultaneous lobing as
illustrated in Figure 3-4. The difference between the signals
received from the two lobes has a null at the Doppler fre-
quency corresponding to the null in the difference radiation
4functions Spattern. The nullreflectivity
of the frequency and the null
slope. directionthearenull
By tracking not
frequency, terrain bias in the heading component is almost
eliminated. The finite tracker bandwidth causes some terrain
bias effect to remain. The reduction in heading velocity
error obtained by lobing techniques is of the order of [2]

6! Vh'/tvh
=Wf2/Af (3-53)

where Af is the Doppler signal bandwidth and Wf is the

lobing low-pass tracker filter bandwidth. The APN-200
Doppler bandwidth at 400 knots is about 550 Hz; while the
tracker bandwidth is 60 Hz, so the error reduction is two
orders of magnitude. Theoretically, further error reduction
could be obtained with a narrower tracker bandwidth. However
this would increase the random error variance as was indicated
by Equation 3-8.
Because of the four-beam symmetry, in level flight the
beam angles of incidence are all 24 degrees. The shifts in
direction cosine magnitudes due to terrain bias are the same
(62's equal, 6m's equal, 6n's equal). The error in velocity
caused by these direction cosine errors is governed by
Equation 3-27, whose error matrix E is

6kk 0 0
E= 0 6m/m 0 (3-54)
00 6n/n

-35-

, ,.h ,.•;• Q Oi(AIr-. /uol C&(O,;GOtD WL;

I, C, GA•A. . ;tU, - ,
•(,i ,.21 .3 17) (. -18,,,
 SIMULTANEOUS LQ13ING 2

Doppler SpectrumG
No Terr-in B~ias

Terrain Rleflect;vity
Causing Terrain Bias 0

(2;) Center of Power- 'ý

o-'- h'-eTerrain Bias

!a±.-Sum Tracker Bandwidth

Doppler Spectrum
with Terrain Bias G

Null point Does Not

Shift Wvith Terrain Ba----
+ -- 1 1 1 --- Difference
Tracker Bandwidth

Correlation
Voltage _____

Tuning F'requency

Fig. 3-4 Simultaneous Lobing11

-36-
INTERMETRICS INCORPORATED -701 CONCORD AVENUE -CAMBRIDGE, MASSACHUSETTS 02138. (617) 661-18
Equation 3-54 shows that in level flight, terrain bias causes
scale factor errors. Teledyne-Ryan estimates that over
smooth water, the heading and drift errors of the APN-200
are [1]

Svh = h .00021 vh
v

(3-55)
6V= - .0202 vd

From Equation 3-54, the equivalent direction cosine errors

are S2 = -. 00007, 6m = -. 00406 radian. The third direction
cosine error from EQuation 3-30 must be Sn = .00089 radian.
So the vertical velocity error over smooth water must be

svz = .00097 vz (3-56)

The values of terrain bias velocity error for three

types of terrain are presented in Table 3-1

"Terrain 'Vh/Vh Svd/vd 6v z/Vz

Smooth water -. 00021 -. 0202 .00097

Smooth barren 00008 -.0079 .00038

terrain
Rough or wooded -. 00001 -. 0014 .00007
terrain

Table 3-1
Terrain Bias Velocity Error

-37-

"G I I C'.. 1 11
 The large over-water lateral error could be compen-
sated in the associated digital computer, provided the com-
puter is told during which portions of the flight the vehicle
is over-water. Some residual error would remain due to
reflectivity variations as a function of sea state and air-
craft pitch and roll. Teledyne-Ryan estimates the residual
error would be about one-fourth of the original uncompen-
sated error[l].
The above analysis assumed the aircraft was in level
flight. In maneuvering flight at non-zero roll and pitch
angles, the angles of incidence of each beam in general are
different. This leads to different direction cosine errors
for each beam. As a result, according to Equation 3-28, the
error matrix is not necessarily diagonal. The Ezd coefficient
would become significant at non-zero roll angles. The Ehd
coefficient could be significant at combined non-zero
roll and pitch angles.

3.6 Surface Motion

A Doppler velocity sensor measures the velocity of the
vehicle relative to the terrain scatterers. Over-water these
scatterers may be moving due to sea or river current or due
to wind-driven surface motion. Since the velocity of interest
is the velocity relative to the fixed earth, surface motion
causes a velocity error.

Major ocean currents, such as the Gulf Stream, have a

maximum surface speed of about 3 knots. Similar strength
currents are observed in coastal areas due to tidal currents.
Throughout most of the oceans, general random sea currents
rarely exceed 0.5 knot.
Surface-wind water-motion error is caused by wind-blown
water particles at and above the surface of the sea.
"Theoretical analyses and actual measurement of this error
have revealed that the velocity error in knots is of the order

3
6v = 1.28 wI1 / (3-57)

where w is the wind velocity in knots.[2] The wind driven

spray and therefore the velocity error is roughly parallel
to the wind.

-38-

IN LHMETHIGS INCORPORAED• 701 CONCORD AVENUE • CAMWRiUuE. MA6SACtiL, L, I -2138• (617) 6 61 -Ir
3.7 Other Doppler Sources of Error
Several other Doppler sources of error z'ce: transmitter
frequency error, receiver sum and difference channel cross
coupling, frequency tracker bias, and speed of light in the
htmosphere. These sources of error are small compared to the
previously discussed errors, so are only briefly discussed in
this section.

3.7.1 Transmitter Frequency

The APN-200 transmitter frequency is uncertain within
+ 2 MHz of the initial setting and may vary another ± 2 M!{z
with temperature. (11 Variation in the transmitter frequency
from the nominal frequency is dir-ictly an error in the assumed
Doppler sensitivity. This causes a scale factor error in all
velocity components. For the 13.3 GHz transmitter frequency
of the APN-200, a 2 MHz frequency error produces scale factor
errors of .00015 in all components. The velocity error
matrix is

[00015 0 0

E 0 .00015 0 (3-58)
L0 0 .00015

3.7.2 Receiver Cross CouplingEl

Coupling between the sum and difference channels in
the receiver of the APN-200 causes a perturbation in the
tracking null point in the s'imultaneous lobing discriminator.
The error ha. been found to be about 0.03 percent. The
resulting heading velocity error is reduced by a factor .707
because the error is the combined effect of the forward and
aft receiving channels, which are statistically independent.
Drift velocity errors are cancelled because drift velocity
is computed as differences of Doppler frequencies processed
in the same receiver channels. The errors do not cancel in
the computation of vz. The vz error is the same as the vh

-39-

•, J...•.)~
, ,:... ', iA ] •- J ,,.,aLa;l(k,•h " " " ...
 error reduced by the ratio of the direction cosines. The
velocity error matrix is

[00021 0 0
E = 0 0 0 (3-59)
0 0 .000081

3.7.3 Frequency Tracker[1]

Frequency tracker errors are specified for the APN-200

as ± 0.05 percent maximum. Assuming a rectangular distribu-
tion, this is a one-sigma error of 0.029 percent. This error
translates directly into a heading velocity scale factor
error. Errors in drift and vertical velocity are practically
cancelled because they are computed as differences of the
single time-shared tracker output. (Recall as discussed in
Section 2.4 that in the APN-200 all Doppler shifts are
positive. The heading velocity is computed from the sum of
Doppler shifts, whereas the drift and vertical velocities are
computed from difference). The velocity error matrix is

00029 0 01
E 0 0 0 (3-50)
0 0 0

3.7.4 Propagation Velocsi

The velocity of the electromagnetic wave in the atmos-
phere varies as a function of temperature, pressure, and
humidity. The velocity at sea level is about 300 parts per
million slower than the vacuum velocity. Error in propagation
velocity causes scale factor errors in the same way as do
frequency drifts. The scale factor error is proportional to
the propagation velocity error in the vicinity of the radar
set, since the locdi wavelength determines the Doppler scale
factor. The APN-200 scale factor is based on the wavelength
at 40,000 feet. At sea level t.he scale factor error would be
.0003. The velocity error matrix at oea level is

-40-
,NP-f(METRICS INCORPORATFD 701 CONCORD AVENUE .CAMbRIOGE, MASSACHUWLI , 02138 (617) %51-I1
I • Ii

.0003 0 0

0 .0003 0 (3-61)
L0 0 • 0003J
L0

Using a standard exponential atmosphere model, this

error could be calibrated to an accuracy of about 50 parts
per million.

3.8 Non-Doopler Attitude Errors

3.8.1 Definitions
To be useful for navigation, the aircraft-referenced
measured velocity components must be transformed to navigation-
frame coordinates (such as an east-north-down set). This
may be indicated by

^n (3-62)

^d
where v is the measured velocity in Doppler (aircraft)
coordinates, Qd is the computed transformation to navigation-
frame from Doppler-frame coordinates and vin is the computed
velocity in navigation frame coordinates. A first order analysis
of the error in the computed velocity 6vn shows that
n=n vd Cd
n v
6vn = d + 6nd (3-63)

That is, in addition to the Doppler equipment and environment

errors giving. rise to the Doppler velocity error in Doppler
coordinates 0vd there is also velocity error due to the
in the attitude transformation 6C11

Of particular interest are the errors in attitude

using a gimballed inertial navigation system (INS) as the
attitude reference. The transformation matrix (Zd to naviga-
tion-frarne fromt Doppler-frame coordinates is the product of
three distinct traiisiorm-it ions.

Cd
n n C d (3-64)

-41-

16 1 l I . .-
 Cn is the transformation to navigation-frame from INS
stable-platform coordinates, CR is the transformation o
stable-platform from INS-base-plate coordinates, and Cr is
the transformation to INS-base-plate from Doppler-base-plate
coordinates. Errors in the measured or computed estimates
of each of these three transformations contribute to the
total attitude transformation error.
Assuming small attitude errors, each computed trans-
formation matrix may be expressed in terms of true trans-
formation matrix and an error matrix as

3. C 3.T (3-65)

where Ejii is a skew-symmetric attitude-error matrix whose

three independent elements are elements of a small-angle
rotation vector in i-frame coordinates. A first order
analysis of the error in the total attitude transformation
shows that

n
6 Cd n d d d
= Cd (Enp + Epb + Ebd) (3-66)

where Enpd is the attitude error matrix associated with the

to-n-fro.m-p transformation but transformed to the Doppler
coordinates d, and the other two attitude error matrices are
similarly defined.
In this report we have been focussing attention on the
velocity errors in Doppler (aircraft) coordinates. The error
in computed velocity given by Equation 3-63 may be transformed
back to Doppler coordinates, and the term expressing the
velocity error due to attitude error is then seen to be of
the form

6v EV (3-67)

where
d Ed d
Enp + pb + Ebd (3-68)

-42-
iNIiT L-. TRICS INCORPORATED - 701 CONCORD AVENUI •CAMUiRIOGE. MAS.. rit. U21,.*- (617) 661-1"
 3.8.2 INS Stable Platform Attitude Error

The first of three transformations in Eq. (3-64) is

the transformation to navigation from platform coordinates.
The inertial navigation computer maintains an estimate of
this transformation, based on a knowledge of the angular
velccity of the navigation frame and a knowledge of the
angular velocity coimmands (gyro torquer commands) to the
gyro-stabilized platform. The error in the computed trans-
formation gives rise to the velocity error matrix (in Eq.
3-68) of the form

dz d
Ednp "Ez 0 ch (3-69)

EdCh (369)0

where Cht Ed, Ez are the three components of the (small angle)
INS attitudc error vector. The sign convention of each
;,mponent is that the attitude error about a given axis is
positive if the platform is rotated positively (by the right
hand rule) about that axis relative to the INS-computed
platform orientation with respect to the navigation frame.

The magnitudes of these attitude errors depend on the

manmer in which the INS is initially aligned, the quality
of the iJvertial instruments (gyros and accelerometers), and
on the flight path trajectory and duration. For a high
quality IRS, aligncd by self-leveling and gyrocompassing,
after several hours of subsonic flight, the horizontal
c.nponents of attitude error will be of the order of one arc
min and the azimuth error will be of the order of several
arc miln.
In an optimal integration of the Doppler system with
an inertial systen, the three INS attitude error components
would be included in the state vector of the Kalman filter.
These error co..on'nts are time varying. The sot of
differential equations governing the I';S error state (including
attitude, velocity, and position errors) may be found in
t:#- literature on inertial navigation system analysis.

-43-
"#T~R,1('ipCS INCORPO:iAYEDO, 701 C(ONCORlD AVENUE C-CAMLDGE. WkASSACHUSETTS 0213.1 - (617) 661-1840
 iI

Pitman[6] provides an introduction. Britting[7] presents

a more general and advanced treatment. Widnall and Grundy[8]
present general INS error models in the modern state space
formulation.

3.8.3 INS Attitude Readout Error

The second transformation in Eq. (3-64) is the trans-
formation to platform from INS-base-plate coordinates. An
estimate of this transformation is computed from the measured
INS gimbal angles. The error in the computed transformation
gives rise to a velocity error matrix of the same form

0 Cz - d]

E|b KCC
d
U - Ep -Eh 0
(3-70)

where Eh, Ed, ez have represent components of the attitude

readout error. A component is positive about an axis if
the base-plate is rotated positively about that axis relative
to the measured base-plate orientation with respect to the
platform.
The gimbal angles are (in a typical INS) measured by
electrical resolvers. The resolver error can be of the
order of several arc min. In the Litton CAINS platform
(utilized in the Holloman AFB CIRIS system), there are
single-speed resolvers for pitch and roll of accuracy 2.5 arc
min lo, and there is an eight-speed resolver for azimuth of
accuracy .75 arc min la. In level flight, these resolver
errors transform into attitude readout components having
one-sigma values

Oh - 2.5 arc ruin .00073 rad

ad - 2.5 arc mir - .00073 rad (3-71)
o2 W .75 arc min * .00022 rad

-44-

INTERMETRWS INCORPORATED 701 CONCORD AVENUE - CAkMDG1E MASSACHUSETTS 02138 - (617) 661 - P
 The corresponding one-sigma values of the elements in the
velocity error matrix are

[~).00022 .00073'
00022
.= 0 0073 (3-72)

00073 .00073

Resolver error is constant at a constai4l aircraft

attitude. But when the aircraft attitude changes, the
resolver error also changes.

An additional suurce of error, which may be considered

part of the INS attitude readout error, is the difference
in alignment between the platfo-wm true horizontal axes and
the accelerometer-defined horizontal axes. The INS iL
leveled during alignment by nulling the outputs of the two
horizontal accelerometers. Hlowever, because of several
physical effects (accelerometer input axis misaligvnment,
accelerometer bias, gravity deflection), the platform aliqnment
that nulls the accelerometer outputs is not truely level.
This bias in platform attitude error, while having no signifi-
cant effect on INS naviciation accuracy, does directly
contribute to INS attitude readout error. In a high quality
inertial system (small accelerometer input axis misaligrnents
and biases) the bias plattorm attitude error should be less
than one arc min, which may be neglected compared with the
pitch and roll resolver errors. For an INS for which the
bias attitude error is not negligible, it stould be noted
that constant platform attitude errors in east-north
coordinates take on? varying values in aircraft coordinates
as heading chnges.

3.8.4 INS-."nnplor e
The last of the three transform-ations in Fq. (3-64)
is the transformation to INS-base-plato from Doppler-baso-plate

-45-
-TERyJETRIC,. :f-,CO'MfPATCD- 701 COjC.M'D AVNUE CA.M iDG2r. MASSACHUETTS 02133- (617) 661-1A0
 coordinates. If both the Doppler and INS base plates are
nominally aligned with the aircraft coordinate axes, this
transformation is nominally the identity matrix. The error
in the assumed transformation gives rise to a velocity
error matrix of the same form
m0 cz -cd

E -Ez:bd 0 ch (3-73)
LCd -ch 0 .

where Ehi Ed, cz here represent components of the

installation alignment error vector. A component is positive
about an axis if the Doppler base-plate is rotated positively
about that axis relative to the nominal or measured Doppler
base-plate orientation with respect to the INS base-plate.
Sources of installation alignment error include
the initial alignment error (on the ground) plus the subse-
quent bending of the aircraft, due to changes in structural,
aerodynamic, and thermal loads. A very rough estimate of
the in-flight installation alignment errors is

Gh = 5 arc min .0015 rad

ad =0 arc rmin .0029 rad (3-74)
az = 2 arc min .0006 rad

The corresponding one-sigma values of the elements in the

velocity error matrix are
r0 .0006 .00291
.0006 0 .0015 (3-75)
_.0029 .0015 0 J

-46-

INTER&AWTRICS NCORPORATED •701 CONCORD AVENLE - CAMBRMDGE. 'AASSACHUSETTS 02138, (617) 661-1'

..................-. .
 At a given flight condition, the INS-to-Doppler
installation alignment error should be constant. But changes
in speed, altitude, or maneuvering loads will change the
-- tructural, aerodynamic, and thermal loads, thereby changing
the alignment error.

3.9 Combined Velocity Error Matrix

In the preceeding sections, the various sources of

Doppler velocity error were discussed. Most sources of
error were shown to contribute to velocity error in a manner
proportional to vehicle velocity according to the matrix
equation

-6v V

6Vh '•hh Ehd E 1z Vh (3--76)

S6vd Edh Edd Edz vd

6V 1- Ez Ezd Ezz V7

The only velocity errors not related to their sources in

the above manner were: l-the noise due to finite beamwidth,
2) .rror aue to tracker tirne constant when the vehicle is
maneuvering, and 3) error due to surface motion.

Table 3-2 sunrnarizes the sources of error which contribute

to velocity error according to Eq. (3-76). The table shows
the one-sigma value of each element of the velocity error
matrix for each source of error. The total velocity error
due to all of these sources of error is also governed by an
error equation of the form Eq. (3-76), where each element of
the velocity error matrix is the sum of the corresponding
elements of the individual error matrices. Assuming all
sources of error are statistically independent, the one-sigma
value of each element in the combined velocity error matrix
is the root sum square of the individual one-sigma values.
These root-sum-square values are given at the bottom of the
table.

-47-

'TERMETRICS INCORPORATED • 701 CONCORD AVENUE * CAMBRIDGE, MASSACHUSETTS 02138 • (617) 661-1840

S. . , , • ,_•.., •• .•t .',

 • I,
I 0
In
0
8

0
8

0 0 0 0. N

~~ In i r I D

N:•, 0 0 0 0 I
4n 0

::o a'!ei; o o a o o
S0 0

00 0 0

;~ N m. 00 V. t
•10 10 8 0 0

i'• i-a
I.-N
a".a o 0 .0 ,
0
G
0 -$4
0 0 00

--
-- 000~ O
0 an 0
41
.- 4 4) 4) 1 0
$4

0aw
0 ma )2fa i
0) IA
*, - .4
r$ W-p 4J j
i 00 to > w
a N 0 0 4)

3 .. 4 N NI

14I
".4 Qt
410 > 14

z
> 0 V z. ;3
- E-4

0. 44 w4.4
o c,. 41 14

4-

oI
0
-48
.... RiPi( RSA N
ECD 701 ON OR A EN E - C M R D E MA S HU ETT 2 3
'4 7 CO R A N AMT- 6 7 6 -
 Any element of the table exceeding one mil (.00100)
is circled, to point out the most significant sources of
error.

Note that INS stable platform attitude error is not

included in the table. It is usually accounted for separately.

3.10 Doppler Error Model Summary

The error 6v in vehicle velocity, measured by the

APN-200 Doppler velocity sensor and resolved into navigation
frame coordinates according to the INS-indicated attitude, may
be summarized by the following equation:

v = E v + Ev + 6v + Sv + w (3-77)

where
SE = combined velocity error matrix (excluding INS
attitude error)

EI= velocity error matrix due to INS attitude error

6v = velocity error due to tracker time constant

-T

6v = velocity error due to surface motion

w = noise due to finite beamwidth

All vectors in Eq. (3-77) are assumed to be expressed in Doppler

antenna coordinates (or equivalently in aircraft coordinates),
for which the three components are in the heading (forward),
drift (right), and vertical (down) directions.
The noise vector w due to finite beamwidth, as discussed
in Section 3.2, has covariance matrix R specified as being
less than

-49-

NTERMETRICS INCORPORATED . 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138 (617) 661-1840
 .016 0 01
R =v 0 .031 0 (3-78)
0 0 .008.
where v is in knots to obtain velocity error variance R in
knots 2 . The noise components are uncorrelated with each
other. Each noise component has a short correlation time,
of the order of the 0.2 sec Doppler output time constant.

The combined velocity error matrix E is of the form

_E E E
hh hd hz

E=: E= Edh Edd Edz (-9

(3-79)

Ezh Ezd Ezz

In level flight over smooth barren terrain, the one-sigma

values of the error matrix elements are, according to Table 3-2

.00140 .00297 .00330'

1.00067 .00823 .00281 (3-80)

L.00303 .00206 .00059J

The elements of the velocity error matrix in level flight
are generally constant or slowly varying. However, the
element Edd, which is dominated by lateral terrain bias, can
shift suddenly if the terrain changes suddenly.
In maneuvering flight, some of the elements of the
velocity error matrix E will have larger one-sigma values
than those given in Eq. (3-80).
The velocity error matrix EI due to inertial navigation
system attitude error is, as was given in Eq. (3-69)

E= -j 0 h (3-81)

cd h
-50-

INTERMETRICS INCORPORATED • 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138• (617) 661-11
 where ch, Cd, cz are the three components of the (smail
angle) INS attitude error vector in aircraft coordinates.
fphe velocity error due to tracker time constant was
discussed in Section 3.3. It is significant only during
maneuvering flight when beam velocity components are changing.

The velocity error due to surface motion was discussed

in Section 3.6. Over water, surface motion may be present
due to current or wind driven surface velocity. When present
(over water) this source of error can be very significant.

3.11 Error Model Simplifications for Kalman Filter Synthesis

To design a Kalman filter for mixing Doppler velocity

measurements with velocity from an inertial navigation system,
the designer must select an appropriate statistical model
for the Doppler velocity errors. It is usually possible to
simplify the complete error model of Eq. (3-77) without causing
significant loss of -accuracy with respect to the theoretically
optimal performance.

Over land, the surface motion term is zero, so may be

discarded. The tracker time constant term is very small
during non-accelerated flight, so may be discarded in cruise
navigation applications.
The velocity error matrix El due to INS attitude error
has three independent elements (Eq. 3-81) which are the three
components of the (small angle) INS attitude error. If the
Kalman filter is supposed to improve the knowledge of the INS
attitude error, then the INS attitude error term must be
retained in the Doppler measurement error model.

The combined velocity error matrix E (excluding INS

attitude error) has nine independent elements, whose one-
sigma values were presented in Eq. (3-80). One could include
all nine elements of the ,rror matrix as state variables in
the Kalman filter. However, it seems rcas7onable that no more
than three sLate variables are required to model the slowly
varying velocity error. The three state variables can be
directly the components of velocity error 6Vh, 6vd, 6vz.

-51-~

";TERMETRICS INCORPORATED •701 CONCORD AVENUE CAMBRIDGE, MASSACHUSETTS 02138 • (617) 661-1840

[gia
 A better choice, however, is the set of non-dimensional
parameters 6 vh/vh, 6vd/vh, 6 Vz/Vh. This latter choice is
motivated by the fact that for small drift velocity and
vertical velocity, the ratio parameters are constant or
slowly varying, independent of speed. The velocity error
variables, on the other hand, can change directly with any
speed change.
The first state variable 6 vh/vh will be essentially
equivalent to the velocity error matrix element Ehh. This
follows from the data in Eq. (3-80), which show one-sigma
values of the three elements contributing to the ahead
velocity error are chh = .00140, Ohd = .00297, and ahz =
.00330. For typical values of drift velocity and vertical
velocity, the contributions of Ehd and Ebz to forward velocity
error will be smaller than the contribution of Ehh. Thus,
forward velocity error is most nearly a scale-factor error.
The second state variable 6vd/vh is not clearly
related to a single element in the velocity error matrix.
If there is small drift velocity, the drift velocity error
will be related to the forward velocity according to the
matrix element Edh. This has the characteristic of an
azimuth alignment error. But if the drift velocity is large,
the dirft velocity error will be related to the drift
velocity according to the matrix element Edd. This is a
lateral scale factor error. From the data in Eq. (3-80),
showing one-sigma values of Odh = .00067 and add = .00823,
the two error elements are of equal significance when drift
velocity is 8% of forward velocity.
The third state variable 6vz/vh will be essentially
equivalent to the velocity matrix element Ezh. The data in
Eq. (3-80) show that for typical values of drift and vertical
velocity, the contributions of the other two elements Ezd
and Ezz are negligible. Thus, the vertical velocity error
is most nearly a pitch aligrment error.
Each of the three state variables can be modeled as
being governed by a first-order Markov process of the form
"1
x -- x +w (3-82)

-52-
INTERMETPICS INCORPORATED 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138 (617) 661-1r
F,

where the time constant T is selected to model the slowly

varying character of the state variable and the white
noise w is assigned a spectral density N appropriate to
maintain the assumed one-sigma amplitude of the state variable.
In some applications, only horizontal navigation is
of interest, in which case the third (vertical) component
of the Doppler velocity may be discarded. Accordingly, in
these applications, only two state variables need be included
in the Kalman filter, representing the slowly varying error
in heading velocity and drift velocity.

-53-

NTERMETRICS INCORPORATED. 701 CONCORD AVENUE • CAMBRIDGE, MASSACI IUSETIS 02138 - (617) 681-1840
 CHAPTER 4

ANALYSIS OF FLIGHT TEST DATA

4.1 Introduction

Performance parameters for modelling the Teledyne-Ryan

APN-200 Doppler velocity sensor were obtained by reducing
flight test data recorded during a single 2-1/2 hour flight
over the Tularosa Basin on 7 August 1973.
The aircraft ground track is sketched in Fiqure 4-1.
It takes-off from Holloman Air Force Base and flies west to
the -i06.35* longitude line, while gaining altitude. It then
turns north and climbs to 32,000 feet. The aircraft turns to
the south-west, and after a couple of minutes, turns to the
southeast and flies back to the -106.35o lonritude. it pro-
ceeds south at this longitude for approximately 8 minutes,
after which it executes a left turn and flies back north at
the same longitude. The aircraft flies two more courses over
this path, after which it returns and lands at Holloman.
During the flight, the Doppler velocity measurements to be
used for data reduction are recorded at the nominal rate of
once/second.

In addition to the APN-200, the aircraft was equiped

with a CAINS baro-inertial navigation system and a CR-100
ranqe/delta-range measuring system. Four transponders
were deployed on the ground for ran(ing purposes: at
Holloman (#4), at Sunspot (#6), at Stallion Army Base (08)
and at Salt Creek (#9) in the White Sands Missile Rance, as
also marked in Figure 4--1. The CAIN,` and CR-100 recorded
outputs were processed by the CIRTS Post-Flight-Processor
(PFP) [5] to generate optimal estimz,tes of the aircraft's
velocity at the Doppler muasurexuent times. These estimates
were subtracted from the Doppler indicated velocities to
generate measurement residuals, which were then statistically
processed to compute estimates of the Doppler error
coefficients.

-55-
S" " " ,,;O ' ... ,' . ... " . .. . . .
 3,.0

3L3J

IaI

ta

Fig. 4-1 Aug 7 Flight Ground Track

-56-

INTERMETRICS INCORPORATED - 701 CONCORD AVENUE - CAMBRIDGE. MASSACHUSETTS 02138 • (617) 661-1P

~ ~ .V- .. ~ ~*'V C~-1k....... . ...

 4.2 Isolation of Doppler Sensor Errors
In order to obtain the Doppler errors, the true air-
craft velocity must be subtracted from the measurements.
The CIRIS PFP contains a two-pass "optimal" smoother, which
provides a velocity refcrence accurate to the order of .1
ft/sec. Figures 4-2 and 4-3 exhibit the smoother-computed
one-sigma uncertainties in the estimated velocity in geo-
graphic axes. If the CAINS inertial system and the CR-100
precision ranging systcm arc performing according to their
performance specifications as accounted for in the optimal
smcother, thcn tne smoatohcr computed uncertainties may be
interpreted as the refcrcncL system accuracy,

The estimated velocity must be transformed to aircraft

axes for comparison wit'l the Dopnler measurements. This
transformation is effected through the CAINS platform gimbal
angles, corrected by the smoother estimated platform attitude
errors, as given by

Cna =Cap CPn Cnc (4-1)

where
a
Ca coordinate transformation matrix from
geographic coords (East, North, Up) to
aircraft axes (Roll, Pitch, Yaw)

a
C =Coordi.nate t~rans f orat-ion matrix fron.
P CAINS platfor. coords (X, Y, Z) to air-
craft axes, through the platform gimbal
i ngles

coordinate transformation matrix from

CAIINS etstimated 9ecographic frame to
platform coordinates, through the
wander azimuth angle

C coordinate traiisform-ation matrix from

n ceographic ceords
cgeoq~aphic coerds, tothroigh
CAI.S estimated
the smoother
plataorm attitude errors

-- 7-.
Trr 7c;!. i', c ,A]PGA1.D 701 CORD~q AVENUE CA .•:IDE•. J.~ASS•ACHU$LTTS 02138 -(617) 6•1-18A0
 r4
oQo * >
04.
.e40

0
• ~~~- ,,>1...,

"V4

-58-
LN,,TERtERI.S WCRPRTED - 701 COCR AVENUE - CA.UBR]O,.,.
DG-.ASA..CHUSJ.ETTS 02138• (617) 661 -1r
33$

_.ta.

>4 >

459

AENUEMASSCHUS$I
1-01CONCPO 0
ITE10_S0236- 7)tZ 1A.
Thus,

cos si n p cosecos* p sine

sin~sinfsiný sinosinOcosp -sin~cosO

Ca +cosocOS -cos4sinOp

cosisinftinp cos~sinOcosýp -cos~cose

-sin cos p +sin~sinp
p
(4-2) J

where
*€ =roll gimbal angle
0 = pitch gimbal angle

9) =azimuth gimbal angle,

p

COSL sina 0

Cp -sins Cosa 0 (4-3)

nC
0 0 1

where
a. wander azimuth angle,

and, muwking small anglo assumptions for the platform attitude

errors

F C:

Cnc -- _ (4-4)
n L - e

-60-

~~~~~~~~~~~~~~~~~~~~~
k) ~R~CC 2~a...nC..~
67
..~G
.v~.
.CiOi 4
..A~~~1.
.iA
.
where
e = platform tip about east

n =platform tip about north

z= platform azimuth error

The platform tips about east and north can be neglected

based on the following considerations: they are consistently
less than 1 minute throughout the flight, and hence their
effect on the desired residuals will be an order of magnitude
below the Doppler induced errors; their effect on the along
and cross heading velocity estimates is through the vertical
velocity, which is well below horizontal speeds. The platform
azimuth error, the largest of the three, and furthermore,
directly incident on the Doppler boresight error, will be in-
cluded in the transformation. Hlence, the resulting transfor-
mation matrix is given Ly

cosOsintp cosOcosI sinO

sin~sin~siný sinsinOcosi -sin~coso (45)

C +cosýco4 -cososinl45

COS€sin0~iin•
iij cos~sinCecosy -cos~zcosQ
-sin~.cosQ sissn

where
ck-- = aircraft heading

The estimated velocity in aircraft axes is then given by

'va = ca ^.
v4-6

where
v"a Actimated velocity in aircraft a4 es
_n
v Ostimatc- velocity in geographic axes

-'; -
 The CAINS indicated velocity contains a deterministic
error resulting from what appear to be out of specification
accelerometer scale factors. This error has been compensated
in the PFP by the following expressions

eve eve + ASPx -fintxcos(a) - ASF~y finty ssin(c)

+ YAZ.fint "sin(c)
x

ev = evn + ASFxf intx*sin(a) + ASFy*f intycos(a)

- YAZ-f. *cos(c) (4-7)

x

where
Sve = error in east velocity (CAINS-true)

ev = error in north velocity (CAINS-true)

f xfnt = X
integral
axis (Xof specific force along platform
accel.)

f ~nt = integral of specific force along platform

y Y axis

L = platform wander angle

and
ASFx = X accE.lerometer scale factor = 4000 ppm

ASFy = Y accelerometer scale factor = -4900 ppm

ASFz = Y accelerometer
platform axis = misaiinment
- 1 min
about Z

-62-
INTERMETRICS INCORPORATED *701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138 • (617) 661-11
 II

The smoothed velocity is provided by the PFP at the

CAINS read times, which are synchronized with doppler
measurements. The doppler residuals are then generated
by differencing the estimates from the doppler velocities
in aircraft coordinates

6A (4-8)

where

6v doppler measurement residuals in aircraft

axes

v-m = measured doppler velocity in aircraft axes

A contributor to the above difference between IMU/PFP

derived velocity and doppler measured velocity is the relative
velocity due to the fact that the doppler antenna is not co-
located with the IYAU. Aircraft attitude rates will induce
additional velocity at the antenna, proportional to the IMU-
to-antenna lever arm, i.e.,

6va = a Ra (4-9)

where

6Va = antenna lever arm induced velocity

-- increments
• a
0a = attitude rate in aircraft axes

Ra = antenna lever arm in aircraft axes

and * .
sinO
q-
a (4"
W = cosq, + icosOsin4 (4-10)

-Osiný + q)cosOcosc

-63-
411 O-F11'ICS INq ." ',..H , ED •701 C-'.ONCO D)AVI NUE , CA ,3 Bf!I[)(.,E, MASSACHIUSETIS 02138 * (617) 661-1 -10
 with
= roll angle

0 pitch angle
4 = heading angle

and () indicates a time derivative.

In the test aircraft, the antenna lever arm was

a 0. (-11)

Hence

(Ocos4 + 4cos~sinl) • 3.58

6v = (-;sin4 + ;cos0cos)• 10.666 - (4 - sin0) 3.58

-(;cos0 + ;cos0sin,)• 10.666

(4-12)

The only persistent attitude rate is the heading rate 4.

For coordinated turns at 30 * bank angle, at a speed of 600
ft/sec, this angular velocity is of the order of .03 rad/sec,
which, at zero pitch angle, induces velocities of

v =
r.0551
.34 ft/sec (4-13)
1.165J

-64-
INTERMETRICS INCORPORATED 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138 • (617) 661-1'
 These components of relative velocity are sufficiently small
(compared to oth°ir sources of Doppler velocity error)
E so that an antenna lever arm correction is not required.

The set of non-dimensional error variables Svh/vh,

6vd/vh, 6Vz/Vh is -elected for statistical analysis. As
discussed in Section 3.11, these variables may be constant
or slowly varying independent of speed. Symbol definitions
lire

6V h along heading doppler measurement residual

6vd = cross heading (drift) doppler measurement

residuai.

6v= along yaw axis doppler measurement residual

and
vh = along heading eoppler velocity

vh is plotted in Fig. 4-4 with flight events marked,

while the doppler residual ratios are plotted in Figs. 4-5,
4-6, and 4-7 respectively. For purposes of readability,
only every fifth available point was plotted. Also, over-
layed on these, is a graph of the sample mean computed
using the nearest 100 samples.

4.3 Segment Selection for Statistical Analysis

In order to obtain a measure of the variability of

the doppler errors statistical parameters, it is desirable
to break up the available flight data into time segments,
each of which is to be reduced separately. The segment
selection is based on the following considerations:

a) Flight Path - the selected segments should

not contain samples pertaining to different
conditions, i.e., samples obtained over
land should not be mixed with samples obtained
over water (if there were any), straight and
level flight portions should not be mixed with
maneuvering segments or with altitude
transitions, etc.

-65-

UC'{..t;t'l{Cj * CORPORATED 701 CONCORD AV[NUE CAMBRIDGE, MASSACUUSETTS 02138 (617) G61-1840
. .. .. ..
 '(22

I
C3 z

W
cc0
2W 4,~ .in

~ .4a

z go
0

c*r

I. a.-t~

c 0)
T r~4

i.. I.

zz

Z ae

-66-

) A~METRICS INCORPORATED 701 CONCORD AVENUE *CAMBRIDGE, MASSACHUSETTS 02138. (617) 661 -1r
 Ia t-

4-4

s-4
__ _ _ _ _ __ _ _ _ _ 4.

C30

NCMOAE 0 CNODAEU - CABIGMSAHSTS018 (1)6114

 I44

S4:

,4'

44

41I

/ __

!0

-68-
INTERMETRICS INCORPORATED" 701 CONCORD AVENUE •CAMBRIDGE, MASSACHUSETTS 02138, (617) 661-1P
 4-)

C50

>1-

4-)
tj H

4-)

CR-
ci C2 c?

ED701
CNCORDAVENUECAMBRI)GE.MAISSCHUSETS 0238 67 6-8
'.!'.!CORPO~l
* t

b) Uncertainty in estimated velocities - the segments

should correspond to those portions of the
flight where the PFP computed 1 sigma
uncertainties in the velocity estimates are
low enough so that they do not represent a
significant portion of the computed measure-
ment residual, and
c) Significant Sample Size - the selected segments
must contain enough samples so that the
uncertainty in the computed statistics is much
smaller than the actual parameter. Thus, for
example, the variance of the sample mean of
k samples is given by
2 x
2< "
(4-14)

where
Gm2 = variance of the sample mean
2 = variance of the individual samples

The flight includes 5 straight and level legs, 3 on a

southerly heading and 2 to the north, of approximate durations
of 8 and 12 minutes respectively. The lack of data points
seen in Figs. 4-4 through 4-7 in the 80150-80400 secs
interval is due to the presence of a data gap in the Flight
Data Tape, so that only about 8 minutes of data are
available for the first northerly leg. Interspersed with
these legs are 8 left turns and 3 right turns at coistant
altitude, all at approximately the same 300 bank angle.
From Figs. 4-2 and 4-3, it can be seen that the
uncertainties in the estimated velocities are of the order
of .06 ft/sec for the horizontal and .11 ft/sec for the
vertical, for the straight and level legs, while climbing
somewhat at the turns. In either case, however, the values

-70-
INTERMETRIGS INCORPORATED' 701 CONCORD AVENUE -CAMCRIDGE. MASSACHUSEITS 02138. (617) 6-IP
 are well beiow the level of the doppler induced velocity
measurement errors, and are therefore acceptable for data
reduction.

At the nominal CAINS readout period of 1 sec, the

straight and level segments yield on the order of 400
samples. Also, from Figs. 4-5, 4-6, an, 4-7, sample
standard deviations of roughly .006, .01 and .003 can
be inferred for the doppler residual ratios along each of
the aircraft axes. Hence, the following uncertainties in
the computed sample means can be expected for this sample
size:

a =0003 along heading

a°MD .0005 cross heading (drift) (4-15)

O mZ =Q0015 along aircraft yaw axis (vertical)

all of which are well below the desired parameter accuracy.

Examination of Figs. 4-5, 4-6, and 4-7 shows signifi-

cant shifts of the mcatn at the reu-ýuverinq instants.
ThereforeR it was decided to compite statistics for these
portions of the flight, even though the available sample
size is significantly smaller than for the straight legs.
The average sample size for the tur-,. is on the order of
50 samples, which yields significance levels of

o =.00085

oM
aD .0014 (4-16)

o mZ • .0004

-71,

I *L~n ~ ' ~ , W .A~C~~ 1 23 I 1C I~

 On this basis, then the following segments were
selected.

1. Straight and level

tstart tstop heading sample size

78950 secs 79400 secs 1800 363

79600 sees 80100 secs 00 418
81000 secs 81500 secs 1800 420
81700 secs 82490 sees 00 657
83000 secs 83550 sees 1800 446

2. Left Turns
a) From north heading to south-west heading.

78360 sees 78430 secs 64

80454 secs 80516 sees 58
82507 secs 82589 jecs 57

b) From south-west heading to south-east heading

7866S sees 78708 sees 28

80753 sees 80793 r-•io 28
82780 sees 8282.b •rrs 28

c) From south heading to north heading

79435 sees 79537 secs 78

81520 sees 81620 secs V

3, Riaht Turns
b) From south-east heading to south heading

78882 secs 78904 sees 21

80950 sees 80975 secs 24
82974 sacs 82994 sees 19

-72-
INIVTEi4.TRICS INCORPORATED - 701 CONCOqD AVLNUE - CAURIRDGE. WASSACHUSETTS 02138 (617) 66£-1P
 After the dzta rcduction process had been completed,
it was discovered that the transponder survey data provided
to Intermetrics, and used by the PFP, was incorrect. In
particular, the a:ssumned survey data for the Stallion
transponder (),represented a location which is 100 ft
above the actual transponder.
The effect of this survey error on the PFP velocity
estimates uas than investigacted. The transponder survey
error is exhibited as a discrepancy between the PFP's
computed range and the actual CR-100 measured range, with
the effect that the estimtated aircraft position is
diL~placed to null the difference. The PB*P assumes an
uncertainty in the measured range due to propagation error,
which is proportional to range. Thus, far away from the
Stallion transponder, the effect of the survey error is
negligible:.

On thu other hand, around the Stallion transponder

overflight, the c~tra:daircraft state is st&-onqly
decpndent on maureaents to that traispeoneler. it can be
shown that for this survey error, the estimated vertical
velocity error will c:xhibit a doublet, -,entered at the
overflight, with a pea'k vitlue of I. to 2 ft/sec; and the
e:Aiiii-tcd aheQad velo~cit~y error will c.,hibit a triplet,
ctnter-ad -itthe overil~ight, with a pteak error of approxi-
tm,,tely 2 it/sec. This ioepeted behavior is confirmed -by
rigs. 4-5 and 4-7.

Hence, for an aircraft speed of about 600 ft/sve, the

errors ill the estim.ate"d ve-lor-ities are of a magnitude
larcier thinn the ex~pvctedJ Ukoploer coefficieants. Therefore,
the~~~~~~ daaetandi
tei ons, i.e., the turns from
north I ading to zin ih.ct rdfo xaiq otws
hezndinq to iut'hest hf(,-dicnq, are not appropriate for
reccovery of doppler narý-smvcrt;. They are neverithele-s-s
includcd in tho follo-iring tabl.es for pu.-poses of complete-

. C-07;osrl-
 4.4 Statistical Analysis
The following statistical parameters are computed as
a measure of the doppler radar errors:
I. Sample Mean, given by

n
i=1

where
n = number of samples
xi = individual sample at time t

II. Standard Deviation, given by

% =~- i(xi M
i)2 (4-18)

"111. Cross Correlation Coefficient, given by

XY -I " (xi - M•.) (Yi .0 (14-191

IV. Auto Correlation Function, given by

(T) 1nri (4-2G)

vhere
ti - t= (4-21)

for discrete values of T(O, 1, 2 secs)

-74-

INTERMETIRC.S -CO,•?ATED -701 COR',,D AVEJE • CALWiDOG. MLASSA.CHUSEIS 02138 - j6tl) 66L1-Ir

- -,:!:•- • :ci.: - "" . z \ -. 4 , " • . • L" " • - ', " • i• " ' :' • : ; •,i : .• • • . , :' ' : : - ' ' -: ' ; :
 gible 4-1 presents sample means and standard deviations
.tc~d from the straight and level segments. Sample
.1 Ind the significance of the sample mean along each axis
.i::o indicated for cach segment. Figs. 4-8 and 4-9
. histogr,-,-s of the Sample means for the ahead and
:t errors. Table 4-2 presents the cross correlation
"...c'n the level-axis errors plus the time correlation of
'.i level axis errors.
tilable 4-3 presents sample means and standard deiv-tions
S-:u~te for the turning segments. Figs. 4-10 and 4-11
-';•inthistogrwois of the sample o.ueans for the ahead and
.:'t errors for these turning segments.

.5 Fl5.qht Test Results Discussion

Donulo

As ex,)ectcdl, the !k'ppler velocity co,[ponents contain

u additive random error. The one-sigma amplitudes of
random euroro-, from Table 4-1, were found to be

0 6v / h= .006
'rv
i/Vh

av d vh 010 (4-22)

6v /Vh
%vz/vh = .003

' " measured values mo-iy b3- co.pirced with noise si-ecification
An Eq.'1ý ",(7.Thairf'et.aveloctyv i11 this flight
ýý--d 355$ C,--vr~tiincýLl a-pe cicati.I olf L-q. (3-78)
.- 4,i.•Ca.oa vzlhr:s .-t 355 knots yield values of .007,
,
1 :nd .005. Fhc M.-ve mvi su.ed .noine- in tihe hr~iriinq and
C, .... • ;reY- to he a4t th -' icati10!1
The mca-*Ircc
• i n the vertical velocity is less
-h n[revi- 1 d level,

-75-
CD .
7k
;RAT C47 c&Ni r I'Ju * I 1M-'t
 Stil 0 I-s 0 .-4
F4 .4 A r4 N
0 0 0 0 0
4J 0 0 0 0 01
.00, 0 0 r., 0

144
0
0 0N
0 A0 tO 0t
P00
04) >'0 0Q 0 00 00
4 00

4141
.4Q4

SN
A r ~O 0 N
N N

+ 0 CO 0 • •0
C!

4) .
060

- 4 4.

0 00 0 0

+p4 to
a
0 N ull .0
*-
' 0

'00 a 9 a
•.0 rd

.1 0 - -4

(O) .0• 0.+. I• 0,4 0

0 0 0 0 0 ?4
S0 00 0 0 00- 0 00

'0 a a 41U a 0g4

0n 0
.0 0> f'
in It) L4 0 a )0 4.J
.n
a 07
L-1 0. 0r. 0~

r-- --A
0 *00000 0 00
>4 ~ a * 0V 104
40 00
HTM0 0O 0C A0 0 C I M
F)~AN Nl N 0'
>->4 o0
>6 0
N- 0 'A4 H
0 0 0 40 w0
s. 0
. % %* to **

go 14 14 1 4 1
+>
a% 0
N o 0 4ý f0

0 0 0q 0 0 U
0 )
c) C. 0 14 '.

.-0 -4 N4 N1 (4 41

4) 4 a ) W

p. (4 0' $4

-76-

INTERMETRICS INCORPORATED. 701 CONCORD AVENUE - CAMBRIDGE, MASSACHUSETTS 02138. (617) 661-18"
 I

MEA -- O,

ju

: L 0
-0.3 -0.2 -0.1 0

DOPPLER AHEAD ERROR|, PER•CEN'T OF AHEA3 VELOCITY

I EN O.D

Fig. 4-8 Histogram of Heading Velocity Er'rors in Level Flight

-77-

'TERMETRICS INCORPORATED.- 701 CONCORD AVENUIE •CAMvBRIDGE, MASS'ACHUSETTS 02138 •(617) 661-1840
 3-
•. MEAN a-0.1 1%
RMS VALUE -. 0.13%,

2 -

4-3 -0.2 -0.1

DOPPLER DRIFT ERROR, PERCENT OF AHEAD
VELOCITY

Fig. 4-9 Histogram of Drift Velocity Errors

in Level Flight

-78-
INERMETRICS INCORPORATED. 701 CONCORD
AVENUE " CAMBRIDGE, MASSACHUSETTS
02138 • (617) 661-18"
 U ~ 0 0. 0~

0 0 0 0 01
H4 H H H
x x x

0. I0

0t . >
0 0 0 0%D.

$4)

4 ; 4 1

oo %

U-' 0 0 0 10 0
H H H H H

H)
C14 HI 0

04-

u 4J
0~ $4

$4 CL$
0t ~ I) t f

II I I E-4

414

cn N v 0

_744
'I~tli''f-I'F%' 01 ON';t~i AV NUF - ( M18 IDGE tA';A('l 11SE TS 0 138 (61) 6 1 -8-1
 * O%1co o 0 0f4
N 000 000 00 0o
H 000 c0o 00 coo
S000 000 00 000

400&f ~
N- r--4#-1 ( FI-1-4 P-4-4
M04 N N
000 000 00 000
30 ;> 000 000 00 000

'0~~ 4-1***

0vd co -4 CID00a*co -1 q.-me-4

WEDO 0o~~ tt- o

-.4

000 000 00 000

c0c0
o Do
c o o 00 ooo
O 4)

I ) A 000 000 00 000

I-
.9N
r- L11r- inL o%00 - n f 0 co
00o0 000 00 000 41
> 000 000 00 000
-4JU
4.,4
td54

00 00 0 04000 a

9 0

.1J cot0
(1 %0 %D) m m rN0 0 %C
>
a)o
A
NOL-W V&A
000
-om0% ~ 000
u
W`
00
lNo 0(
ooot-
0 o00
0:
V
9
000 000 00 000 1.0

>4 54.45

.00

:0 M'404% 0- -0 r 0 ý
1 O 0 P-
N,An twiou0,1 r-4 (n
.0 00o0 000 c0 000
0 '0000. 000 E E-00
60 %

M'TI.
ICOPO
44
*NT TE
)
OCR .70
10
*AEU . CABIDE *,&SCUET 021 0 67 611
01 .04

04' 'O9r to 4'-4O

4 A oo00 00 000 44.:

z A 0( '00
00 Soso0 go

OW
A0
w' 41kV.I0 -

INTLMI*RIC 0 N In 14 N
INOPRTD71 OCR VNE ABIGOSSCUET 238(1)611

0 A0l
7: I

LrLI
o

W s-
*4J

I
H

S> 0

w ~14

0z 0

ro

_____ ____
_ __ __ ____ -
eq N- .543
H
-8K
MA.

C4.

W P4

TFRMVETRICS INCORPORATED. 701 CONCORD AVENUE *CAMBRIDGE, MASSACHUSETTS 02138. (617) 661-1840
 *1

.6-I

< ~4-
r4

C#

o L•
Wu >1

c4.)
I- 0
Lu o

-. l

100
am-

-82-

INTERMETRICS INCORPORATED. 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138 •(617) 661-18"
 In theory, the cross-correlation between the
components of Doppler velocity random error should be
zero. As discussed in Section 3.2.3, this can be shown
assuming independent random error in each beam velocity
and by taking into account the four-beam-array symmetry.
The measured correlation coefficient between the heading
velocity random error and the drift velocity random error
was shown in Table 4-2. It is seen to be small, in
agreement with theory.
In theory the Doppler velocity random error should
have a very short correlation time, of the order of the
0.2 sec Doppler output time constant. The measured
autocorrelations of the heading and drift components of the
Doppler error were presented in Table 4-2. The values of
correlation at T = 1 and T = 2 sec are seen to be generally
an order of magnitude below the values at T = 0. These
results are consistent with the correlation time being less
than one sec.

4.5.2 Slowly Varying Errors in Level Flight

During the five segments of straight and level
flight, the ahead velocity error variable was found to
have segment mean values ranging from -. 0008 to -. 0012
(Table 4-1). The average value for all segments was -. 0010.
According to the Doppler error model summarized in Section 3.10
and in particular according to the expccted one-sigma values
for elements of the velocity error matrix (Eq. 3-80), for
small drift and vertical velocity the ahead error is expected
to be about .0014. The measured ahead error is found to
be slightly smallkr than expected.
The drift velocity error (Table 4-1) was found to vary
from -. 0003 to -. 0024. The average value for all five
segments was -. 0011. According to the Doppler error model
summarized in Section 3.10, the non-dimensional drift error
variable is dominated by the contributions of the two
velocity error matrix elements Edh and Edd

6 vd/vh = "dh + Edd Vd/Vh (4-23)

-83-
*rfUp•I jtjS INGOP'OR•AI~t) •701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 0?138 • (617) 661-1840
 These elements according to Eq. (3-80) are expected to have
one-sigma values of .00067 and .00823, respectively. During
the northerly and southerly legs of the flight there was a
crosswind of about 60 ft/sac, which produced an equal
magnitude drift velocity whose sign depended on the flight
direction (north or south). The heading velocity was 600
ft/sec. For the drift to heading velocity ratio of 0.1,
the second term in Eq. (4-23) is expected to have one-sigma
value of .00082. The combination of the two terms having
one sigma values of .00067 and .00082 with the sign of the
larger term alternating with flight direction would produce
a drift velocity error having large shifts in value with
flight direction. This characteristic is indeed observed
in the measured drift velocity error in Table 4-1. The
data is biased (mean value -. 0011) more than is expected
considering the one sigma value of Edh of .00067. Some
contributor to Edh is larger than expected. Perhaps the
Doppler-with-respect-to-INS installation azimuth alignment
is larger than the assumed .0006 radians (2 arc min).
The vertical velocity error (Table 4-1) was found
to vary from .0002 to .0025. The average value for all
five segments was .0013. According to the Doppler error
model summarized in Section 3.10, for normally small drift
and vertical velocities, the vertical velocity error should
be dominated by the element Ezh which is expected to have
one-sigma value of .00303. The measured vertical velocity
error is below the expected one sigma value, indicating
that the velocity error matrix element Ezh may be smaller
than expected. The largest contributor to Ezh was assumed
to be the Doppler-with-respect-to-INS installation pitch
alignment error of 10 arc min one sigma. Note that the
measured vertical velocity error shifts significantly with
heading. Such behavior could be caused by a steady INS
readout error caused by steady platform tip about east of
the order of .0008 radian (3 arc min). Such a large steady
tip could be caused by an 800 Pg north accelerometer bias
or a 3 arc min north accelerometer misalignment about the
platform east axis.

4.5.3 Slowly Varying Errors in Maneuvering Flight

As ir,:cated in Section 4.3, significant shifts in
the mean v.. ty errors are observed during the turns.
Thcrefor,- se-rate sample means were computed for these

-84-
INTERMETRICS INCORPORATED 701 CONCORD AVENUE .CAMBRIDGE. MASSACHUSETTS 02138 , (617) 661-11
 segmaoi.t:; ad were presented in Table 4-3. The tabulated
results l-ow component velocity errors several times
larger than those for level flight.

As ij-entioned earlier, the sets of data near trans-

ponder 8 should be disregarded, because an error in the
assumenxd altitude of the transponder (survey data) produced
significant velocity error in the best-estimate-of velocity
as obtained by the CIRIS Post Flight Processor (PFP).

Ain additional source of reference error was discovered,

but found to be negligible. A wrong value was used in the
PFP for the CR-100 -anging system antenna lever arm. The
given roll axis component was too large by the order of 4 ft.
Based on similar considerations as for the Doppler antenna
lever arm, this effect does not contribute more than 0.1 ft/
sec and hence is negligible.
Several Doppler-velocity-measurement sources of
error are expcvcted to have varyinq and perhaps larger values
during nraneuvcrs. Some of these in the velocity error
matrix are the lateral terrain bias, the INS attitude read-
out (resolver) error, anrd the INS-Doppler installation
alignment (if there is a significant mareuver g level). In
addition the effect of velocity lag due to Doppler time
constanit can become significant.
However, it is also true that the reference system
velocity errors may also be larger during maneuvers. The
CIRIS reerence system, at the time of this 7 August 1973
flight, wzas just reaching operational status. As with any
Snew s,,Jm, it is likely that a few bugs were not yet
discovcered and corrected. The transponder 8 survey error
is on- error that has subsequently been discovered and
corryct;V.- 'ie indicated velocity from the CAINS inertia,
1yt1;1i ES1hifLs in apparent velocity error during turn,•
as -if `-orc' 1:ere horizontal acceleromuter scale factor
error:; c[ AS;I. -0;) 0 ppm and ASr'y= -4900 ppm. This observ,:-
sys( .!, (-error in the INS velocity was compensated in tý-
Po.;L "1i 111t "'rocossor, as indicated by Eq. (4-7). Howev,::f*
wf% c-Ji:f't be certain that this error (whatever its sour(e)
ha:.; i: v.,ltilely eliminated. Therefore the possibility

-85-

• " 1 '" ',"`'CORD AV-NU? - CAMBRIDO(', MASSACHUSETTS 02138. (817) 661-1840

 exists that the CIRIS PFP indicated velocity is not as
accurate as the PFP-computed uncertainties presented in
Fig. 4-2 and 4-3.
Accordingly we hesitate to conclude that the measured
velocity differences for maneuvering flight are due to
degraded Doppler performance during these maneuvers.

-86-
:*ERiCS PCORpORATED. 701 CONCORD AVENUE - CAMWBFDGE, MASSACHUSETTS 0213& - (617) 661-IP
 CHAPTER 5
SUMIARY AND CONCLUSIONS

A Teledync-Ryan APN-200 Doppler velocity sensor has

been added to the 1iolloman Air Force Base CIRIS (Completely
Intcqrated Reference Instrumentation System). The major
CIRIS svinsors include a Litton C7INS inertial navigator
with barometric altimeter, a Cubic Corp. CR-100 precision
ranging systcm, and now the APN-200 Vloppler. Intermetrics
was asked to modifN the CIRIS Post-Flight Processor to
incorpor ate the Doppler mea!surements into the optimal
estimation of the aircraft navigation state (position,
velocity, ittitude). This requires an error model representing
the sources of error in the APN-200 Doppler velocity sensor.
This re),-ort provicies such an error mozlel. Also flight test
data have baen analyze- to confirm the statistical parameters
in the (_Zror model.

A birief discussion of Doppler sensor principles is

presented, as well as a dis,-ussion of the specific functional
design of the APIN-200. This introduction makes the report
self contained, by pr-vidinq the necessary bachground for
the discussion of the indAivdual sources of error. Performance
limits that are impliW- by the functional design are
discussed.
The individual sources of error in a Doppler velocity
measurement were discussed, including estimates of their
one-sigqi4 valuos. Manv of the sources of error arc a function
of the dý.iqin of the Doppler velocity sensor. In addition,
several etcrnal environment factors are sources of error.
Also inclv6-d in the error dis'ussion is the velocity error
due to attitude rcference eraor.

Doppler velocty error is shown to -,ontain a large

amplitud. railuom error in all three cormpgintu. However,
because of its short correlation tOie, after processing many

-87-

'i i X.T!•:,. . • ,0; (OrCOti) AV'A:41J"

.... S ,,,: i,....:. e'',. M.U,1) MASSAC.HU;SETTS O? I338
CA',41iIIDYE.K, (6171 6i61. I..o
 Doppler measurements the effect of this zero-mean random
error is reduced to a negligible level.
The more significant Doppler errors are the lower
level constant or slowly varying errors. Many of the
constant or slowly varying sources of error contribute to
velocity error as elements of a 3 by 3 velocity-error
matrix (velocity error being the product of this matrix
times the velocity vector). For the assumed one-sigma
values of the APN-200 and other sources of error, and
assuming __nderatu or small values of lateral (drift) velocity
and vertical velocity, the most significant elements of this
matrix are Ehhj, Edh, Edd, and Ezh.
Euhh is the element relating ahead velocity error to
ahead velocity (scale factor error). The largest contributor
to this element is estimated to be the Doppler beam direction
error due to the antenna being at a temperaturc different
from that at which it was calibrated.
Edh is the element relating drift velocity error to
ahead velocity (azimuth alignment error). The largest
contributor to this element is estimated to be the alignment
error between the INS inside the aircraft and the Doppler
antenna mounted or the outside of the aircraft.
Edd is the element relating drift velocity error to
drift velocity (lateral scale factor error). The largest
contributor to to this element is estimated to be the terrain
bias effect over the smooth barren terrain of the Tularosa
Basis in New Mexico. Over smooth water the terrain bias
effect would be larger. Over rough or wooded terrain the
effect would be less.
Ezh is the element relating vertical velocity error
to ahead velocity (pitch aliqnment error). The largest
contributor to this element is estimated to be the alignment
error between the INS and the Doppler antenna.
A brief discussion wav presented on how the complete
error model may be simplified for the purpose of Kalman
filter synthesis. Three state variables may be used to
represent the Doppler velocity measurement slowly varying
error in each velocity component.

-88-
INTE•METrICS INCORPORATED - 701 CONCORD AVENUE - CAM1RDGE, MASSACHUSETTS 02138 - (617) 661-IP
 The data from a flight test of the APN-200 were
analyzed. The Doppler Tmieasured velocity was compare.d with
an estimate of vehicle velocity provided by the optimal
combination of the CIRIS inertzial and precision ranging
data. The aptimal processor has a self-estimate that its
velocity errors ,ere lezs than 0.1 ft/sec for east and
north velocity components and less than 0.2 ft/sec for the
vertical velocity component. Hlowever at this oarly stage
in the debu-qincl of the CIRIS system. it is possible that
unmodeled soArces of error (in the inertial system, in the
precision ranging system, in the flight software, or iii the
optimal Post Plight Processor) may have caused reference
velocity errors larger than the above values. Thus the
caputed difftrences hetweqn the -aeasured APN-200 Doppler
ve~locity and the estimated refercn•ce velocity may be an
indi.atijon not only of the Dopphvzr errors but also the
refe•rence errors. All that can be concluded with certainty
is thi•t the Dopplct errors must 1% less than or equal to the
observed dx f terences.
The anziy:iz of thr- vcloci. -differenres Thowed the
large zazplitueo xz.ndon! '-iror to be at the specification level
in the ahead and latevp components and to be less than the
specification levl in Cie vertical component.
The constant or slowaly varying portions of the velocity
differenccs •nlevel flight were at the expected level fnr
Poppler forta-d vclocIty error 4nd for Doppler lateral
velocity error. r,.e v etical t-olocity difference ttas some-
whalt smaller than the erpected one-sigma value for &appler
vertical velccity error.
In maneuvciAn fqiMht, the obtsecrved slowly-varying
portions of the velo•i. diifaences were several times
larqoa thanthe LLOinc er'rens i;i level flialt. S&veral
sources of error are knot'nl to pxcrdo incros-rd Doppler
velvcity eirror during ntneuvc-,rs-. lowevcr, th. reference
velocity may also have hd ia-cr-eucd velocity orzor during
the ritaneuverV. Therefore no firn, conclusions can be drawn
frcm this l'ira-
2 ied data with rcpegpcc to the ArN-1200 error
c.'Mracteristics duriw; maneuvers.

7-019-
Th¶L',Tpr" iCfP%1r 0 OY';AgU -. vtJ MA tF~T ?' :~ei t
 VT

Previous flight tests of Doppler velocity sensors

have generally performed a time integration of the Doppler
"velocity measurements and compared the indicated position
with a reference position. The flight test data reduction
in this report has directly compared the Doppler velocity
outputs with a precision velocity reference trajectory.
This approach permits examination of the Doppler random
error as well as examination of the slowly varying error
as a function of flight conditions, such as heading, cross-
wind, and attitude. This approach has the greatest promise
of exposing the individual .:ources of error that are making
the largest contribution to Doppler velocity error.

i I

-90-
ý'-TERMETRICS INCORPORATED • 701 CONCORD AVENUE • CAMBRIDGE, MASSACHUSETTS 02138 - (617) 661-18'
 REFERENCES

1. Anon., "Proposal: AN/APN-200 Doppler Velocity Sensor

for Long Range Patrol Aircraft", TRA 29070-17c,
Teledyne Ryan, San Diego, Cal., January 1973.

2. Pried, Walter R., "Doppler Navigation", Chapter 6 of

Avionic NaviqatLon Systens, M. Kayton and W. R. Fried,
Sdi ftors, John Wiley & Sons, New York 1969.

3. VanderStoep, D.R., Call, R.W., and Warzynski, R.R.,

"A New Approach to Doppler-Inertial Navigation
(Doppler Beam Sampling)", presented at the 6th
V! Asilomar Conference on Circuits and Systems,
Monterey, California, 15-J7 November 1972.

4. Berger, F.B., "The Nature of Doppler Velocity

Measurement", IRE Transactions on Aeronautical
and Navigational Electi-cnics, Vol. 4, September 1957,
pp. 103-112.0

5. Widnall, W. S., Carlson, N.A., Grundy, P.A., "Post-

Flight Processor for CIRIS", TR 16-72, Intermetrics,
Cambridge, Mass., 24 November 1972.

6. Pitman, G.R., editor, et al, Inertial Guidance, John

Wiley and Sons, New York, l•62.

7. Britting, K.R., Inertial Navigation System Analysis,

Wiley-Interscience, 1971.

8. Widnall, W.S., and Grundy, P.A., "Inertial Navigation

System Error Models", TR 03-73, Intermetrics,
Cambridge, Mass., 11 May 1973.

,t i'" ... I..., O,. :D 701 CQNCOND AV,E ' CAM[IRIDGE, MASSACHUSETTS 02138 " (617) 661
•.L p.i,,10