Upload
others
View
7
Download
0
Embed Size (px)
Citation preview
NOAA Technical Report NOS 107 C&GS 3
Algorithms For Confidence Circles and Ellipses Wayne E. Hoover
Charting and Geodetic Services Rockville, MD September 1984
U. S. DEPARTMENT OF COMMERCE Mdodm Bddrlgl, SocmtUy
National Oceenlc and Atmospheric Admlnlstrstlon Anthony J. Wio, Acthg Admhrshator
National Ocean Sew Paul M. Wdfl. Assistant ~ i s t m t w .
NOAA Technical Report NOS 607 C&GS 3
Algorithms For Confidence Circles and Ellipses Rockville, MD September 1984
U. S. DEPARTMENT OF COMMERCE National Oceanlc and Atmospheric Administration National Ocean Service
ERRATA
Hoover, h y n e E., "Algorithas for Confidence Circles m d Ellipses b'
Urshington D.C. , Nrtionrl Ocernic and Atmospheric Administr~tiOlle M O M Technical Report NOS 107 C&GS 3, (1984) pp 1-29.
l e Page 8, In the equation for p(R), Insert a closing brrcltct between the . superscript 2 and the drdy,
2. Page 13, Equrtian 7b incerrcctly reads!
p = fL + T1 + T2 + T31c/n
me correct formula ( f o r the trapesoidal'rulc)' Is:
p z k T l + T2 + T31c/a
3. Page 25, Line 19 erroneously contains "trots" the correct spelling is nerrorsn.
3-27-86 UAYNE E. HOOVER
ALtiORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
Abstract
I n many hydrographic surveylng, navlgatlon, and pos l t lon loca t lon systems
the observed pos l t ion i s defined as t h e in te rsec t ion o f two l i n e s o f position,
each o f which may be i n error. This paper gives algorithms with new stopplng
c r i t e r i a f o r the determination o f t he p robab i l i t y that the t r u e pos i t i on T
l i e s within a c i r c l e o f given radius centered a t t he observed pos i t ion 0, and
conversely, the determination o f the radius o f a c i r c l e C w i th center 0 such
t h a t the p robab i l i t y i s p t h a t T l i e s within C. I n e i the r case, the c i r c l e
centered a t 0 i s ca l led a confidence c i rc le .
Confidence e l l i pses a r e a l so considered and a re shown t o be superior t o
confidence c i r c l e s since they provide the same p robab i l i t y o f loca t ion but
general ly over a s i g n i f i c a n t l y smaller region.
It i s assumed that the er ro rs associated with the l i n e s o f pos l t i on ray
be approximated by a nonorthogonal b i va r ia te dependent Gaussian d i s t r i b u t i o n
where the errors are measured orthogonally t o t h e l i n e s ob position. The
a l g o r i t h m given are s t ra ight forward and easy t o inplement on a microcomputer.
Biographical Sketch o f Wayne E. Hoover
Wayne E. Hoover i s a systems analyst w i t h the National Oceanic and
Atmospheric Administration i n Woods Hole, Massachusetts, and a lso i s an
adjunct professor o f mathematics a t Cape Cod C o m n i t y College i n West
Barnstable.
S t a t e Uni vers i ty . I n 1977 he received h i s Ph.D. i n nunerical analysis from Michigan
-i-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
TABLE OF CONTENTS
ABSTRACT
TABLE OF CONTENTS
ILLUSTRATIONS
TABLES
KEY TO SYMBOLS
1 INTRODUCTION
2 MATHEMATICAL CONSIDERATIONS
2.1 Geometry 2.2 Assurrptions 2.3 Transformation t o an Orthogonal System 2.4 The Error E l l i pse 2.5 Confidence E l l ipses 2.6 Confidence Circ les 2.7 Numerical Quadrature
3 ALGORITHMS
3.1 Algor i thn 1 f o r p(R) 3.2 Algor i thn 2 f o r R(p)
4 NUMERICAL EXAWLES
4.1 Example 1 4.2 Example 2 4.3 Example 3 4.4 Exanple 4
5 EQUIVALENT FORMJLAS FOR THE ERROR ELLIPSE
5.1 Special Cases
6 APPLICATION TO LORAN-C
7 CONCLUSIONS AND RECOMMENDATIONS
APPENDIX: CIRCULAR ERROR PROBABILITIES
REFERENCES
Page
i
li
iii
iii
i v
1
2
2 3
11
12 14
14
14 18 18 22
23
23
25
25
27
28
-i i-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
ILLUSTRATIONS
Figure 1 Nonorthogonal Coordinate System
Figure 2 Orthogonal Coordinate System
Figure 3 The Error E l l i pse
Figure 4 E l l i p t i c a l Scale Factor vs. Probab i l i t y
Figure 5 The Error E l l i pse .. .
Figure 6 Confidence C i rc les w i t h a1 = 2, o2 = 1, a = 30",
Figure 7 Radius o f the 95% Confidence C i r c l e when
andpli = 0
u1 = u = 1 and p l Z - 0 2
TABLES
Table 1
Table 2
Table 3
Parameters associated wi th al = 2, o2 = l r a = 30'.
Parameters associated wi th al = a2 = 1 and p12 = 0
Areas o f 95% Confidence C i rc les and E l l i pses
andpI2 = 0
when o1 = a2 = 1 and p12 = 0
Paqe
2
4
6
8
15
17
2 1
16
89
20
-iii-
C
C
CEP
E
. g i
h
K
K i
k
n
0
P
ALGORITHMS FOR CONFIUENCE CIRCLES AND ELLIPSES
KEY TO SYMBOLS
Confidence c i r c l e
Rat io o f smaller t o la rger standard deviation: o/ax
Circu la r e r r o r probable = radius o f t he 50% confidence c i r c l e
Probab i l i t y e r r o r estimate
2.7182818284590452. . . Integrand
Aux i l ia ry funct ion
In tegrat ion step s i ze
Ratio: R/ax
i t h i t e r a t e o f K
E l l i p t i c a l scale fac to r
Length o f semimajor ax is o f confidence e l l i p s e
Length o f semiminor ax is o f confidence e l l i p s e
F i r s t l i n e o f pos i t i on
Second l i n e o f pos i t i on
Line o f pos i t i on
Pos i t i ve i nteger
Observed pos i t ion
Probabi 1 i ty associ ated w i th confidence c i r c l e o r e l l i p s e
Probab i l i t y as a func t ion o f radius o f confidence c i r c l e
C i r cu l a r e r r o r probabi 1 i ty
Radius o f confidence c i r c l e
Page
8
8
16
11, 13
7 .
9
14
10, 13
8
14
7
7
7
2
2
1
10, 13
1
7, a, 13
a, 13
9
8, 14
- 1 V -
ALGORITHHS FOR CONFIDENCE CIRCLES AND ELLIPSES
X
Y
a
8
e
n
912
p Y
O X
O Y
0
1 dRMS
2dRMS
Radius o f Confidence c i r c l e
True pos i t ion
End point trapezoidal sun
as a funct ion o f p robab i l l t y
I n t e r i o r end point trapezoidal sum
Centroid or midpoint sun
F I r s t nonorthogonal ax is
Second nonorthogonal ax is
Auxi l iary function
F i r s t orthogonal axis
Second orthogonal ax1 s . Angle o f crossing, 0 < a < n 2cAI
(K/2c?
Orientat ion o f semlmajor ax is o f e r r o r e l l i p s e w i t h respect t o L,
3.1415926535897932.
Correlat ion coe f f i c i en t i n ul-u2 coordinate system
Transformed co r re la t i on coe f f i c i en t
Standard deviat ion associated w i t h L,
Standard deviat ion associated w i t h L,
Length o f semimajor axis o f e r r o r e l l i p s e
Length o f semiminor ax is o f e r r o r e l l i p s e
Variable o f in tegrat ion
Radial or root mean square e r r o r
Upper bound f o r radius o f 95% confidence c i r c l e
Page
11, 14
1
13
13
13
2
2
9
3
3
2
9
9
3, 12
a 3
6
3
3
5
5
9
15
15
- V -
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
1.0 INTRODUCTION
Hydrographic surveyors, navigators, and others concesned with pos i t i on
loca t ion have t r a d i t i o n a l l y determined t h e i r pos i t i on by means of two
in te rsec t ing l i n e s o f pos i t ion (LOPS). The LOPs may be derived from c e l e s t i a l
observations, t r i l a t e r a t i o n , LORAN signals, s a t e l l i t e signals, etc.
Two questions important t o pos i t i on locators are the following: (1) What
i s the probab i l i t y tha t the t r u e pos i t i on T, which i s generally unknkn, 1s
located R un i t s o r less from the observed pos i t i on 0; and conversely, (2) Nhat
i s the radius o f the c i r c l e C centered a t 0 such t h a t the p robab i l i t y i s p
t h a t T l i e s w i th in C.
'. I n e i t h e r case, a c i r c l e o f radius R which i s centered a t the observed
pos i t i on 0 i s ca l l ed a confidence c i r c le .
uncertainty o r c i r c l e o f equivalent p robab i l i t y .
It i s a lso ca l l ed a c i r c l e o f
This paper w i l l ou t l i ne the mathematical aspects o f these problems and
then give new algorithms f o r t h e f r solution. The algorithms are s t ra igh t -
forward, e f f i c i e n t , and read i l y implemented on a microcomputer.
Also, mention w i l l be made o f confidence e l l i p s e s which are ac tua l l y arch
easier t o ca lcu late than confidence c i r c les ; moreover, they are superior t o
confidence c i r c l e s since they provide t h e same p robab i l i t y o f loca t ion over a
generally s i g n i f i c a n t l y smaller area.
F ina l l y , several nunerical examples i l l u s t r a t i n g the appl icat ion o f t h e
a lgo r i thms w i 11 be presented.
-1-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES . .
2.0 MATHEMATICAL CONSIDERATIONS
. 2.1 Geometry
Designate the two l l n e s of pos i t ion by 1, and L,, respectively, and l e t
a, 0 < a < n, be the crossing angle from L, measured i n a pos i t i ve O r
counterclockwlse d i rec t l on t o L,. Let 0 denote the In te rsec t ion of the
LOPS. Thus 0 represents the observed or measured posltlon.
.
Uefine the nonorthogonal u1-u2 coordinate system such tha t u1 and u2
In tersect at 0, u, I s perpendicular t o L,, u2 i s perpendicular t o L,, and t h e
pos i t i ve angle from u1 t o up i s n + 0. This geometry fo l lows tha t o f Swanson
[9] and i s I l l u s t r a t e d i n FIgure 1.
"I
"2
Figure 1 Nonorthogonal Coordinate System
Note that Burt, Kaplan, and Keenly, et a1 141 and also 'Bowcitch [2] use a
d i f f e r e n t geometry by reversing the d i rec t ion of the u2-axis. I n t h i s case,
the pos i t i ve angle from u1 t o up equals a. Moreover, t h i s changes the sign o f
t h e cor re la t ion coe f f i c i en t plp .
- 2-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
2.2 Assumptions
Throughout t h i s paper we w i l l assune the following:
I n a small region G containing 0, t h e ear th i s f l a t and the two LOPs are s t ra igh t l ines.
The errors i n the measurements which determine Ill and L, are
normally d i s t r i bu ted random var iables wi th cor re la t ion coe f f i c i en t
zero means, and standard deviat ions o1 and 02D respectively, p12 * where u i s measured along ui which is perpendicular t o Li . The b i va r ia te e r r o r d i s t r i b u t i o n i s constant throughout the region G.
Thus i t i s assumed t h a t the er ro rs i n the measurements o f the LOPs, which
may o r may not be correlated, may be approximated by a nonorthogonal b i va r ia te
dependent Gaussian d is t r ibu t ion .
This paper applies only t o those pos i t i on loca t ion systems f o r which the
It can be above three assumptions provide the basis f o r a v a l i d e r r o r model.
a sizeable task t o decide whether t h i s model i s appropriate f o r a spec i f i c
pos i t ion loca t ion system.
2.3 Transformation t o an Orthoqonal System
Now transform the nonorthogonal ul-% system t o an orthogonal x-y
Cartesian coordinate system centered a t 0 and or iented such t h a t the angle
from 4 t o the pos i t i ve x-axis i s given by 8. Following convention, a
pos i t i ve angle i s measured i n a counterclockwise direct ion. These coordinate
systems are i l l u s t r a t e d i n Figure 2.
The transformation i s given by
5 = xsin(8 ) + ycos(8 )
- 3 -
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
= xsin(a - e) - ycos(a - e). "2
The angle 8, which i s defined i n the next section, i s chosen so tha t the
transformed variables are s tochast ica l ly independent.
Y
0 0
Figure 2 Orthogonal Coordinate System
2.4 The Er ror E l l i p s e
I n order t o determine the radius R(p) o f the confidence c i r c l e C, or the
probab i l i t y p(R) associated with C, i t i s necessary t o f i r s t ca lcu late the
parameters o f the error e l l ipse , namely, the lengths o f t he semimajor and
semiminor axes and t h e i r o r ien ta t ion w i th respect t o a coordinate System.
Def in ing the anc i l l a ry var iables 2
a1 = olsin(2a) + 2p12alu2sin(a)
a2 = a1cos(2a) 2 + ~ P ~ ~ ~ ~ U ~ C O S ( ~ ) + a2 2
- 4 -
ALtiORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
2 5 = 2s in (a)
and using the transformation given i n Section 2.3, i t can be shown tha t the
given nonorthogonal standard dev ia t ions a1 and a2 are transformed t o ox and
respectively, i n the orthogonal x-y Cartesian coordinate system, where O Y S
u = [(a, - q,)/a51 4- Y
and a x ’ L o holds for a l l v a l i d values o f the input variables: a, 20,
a2 2 0, 0 < a < It, and -1 < p12 < 1. The e r r o r e l l i p s e i s the e l l i p s e wi th
center 0, semimajor ax is a x which coincides w i th the pos i t i ve x-axis, and
semiminor ax is oy which coincides wi th the pos i t i ve y-axis.
Y
The or ientat ion o f the e r r o r e l l i p s e i s calculated from
tan (a) = a,/%.
Note tha t t h i s ca lcu lat ion must be performed so tha t 0 i s obtained i n the
proper quadrant. This can be achieved with the a i d o f the double argument
arctangent function o r the rectangular-to-polar function. Thus, -n/2 < 0 <
n/2, where0 i s the angle from L, t o the pos i t i ve x-axis. As before, a
pos i t i ve angle represents a counterclockwise direct ion.
ALtiOHITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
Y
"2 0
Figure 3 The Error E l l i p s e
The error e l l i p s e with parameters ox, u and 8 i s i l l u s t r a t e d i n Flgure 3.
XY
Y ' I n the x-y coordinate system, the cor re la t ion p
var iables i s zero.
between t h e transformed
- 6 -
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
2 a5 Confidence E l 1 ipses
A confidence e l l i p s e i s an e l l i p s e which i s concentric t o the e r ro r
e l l i p s e and which has parameters tu,, by, and 8 ; k i s ca l led the e l l i p t i c a l
scale factor.
Since u and uy represent the standard deviations o f s tochast ica l ly
independent random variables, the addi t ion theorem f o r the chi-square
d i s t r i b u t i o n may be used t o show t h a t the probab i l i t y associated w i th a
confidence e l l i p s e i s given by
Conversely, the semimajor ku, and semiminor ka axes o f a confidence
e l l i p s e having specif ied p robab i l i t y p may be calculated from ax, uy, and Y
L k = [ -2* ln( l - p)J2m
Thus the e r ro r e l l i p s e i s a confidence e l l i p s e w i th e l l i p t i c a l scale
fac to r k = 1 and p robab i l i t y approximately p = 0.3935. The 50X and 95%
confidence e l l i pses have e l l i p t i c a l scale factors approximately l a 1 7 7 4 and
2.4477, respectively.
Figure 4 contains a graph of the e l l i p t i c a l scale factor as a function
o f probabi 15 tym
- 7 -
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
ELLIPTICAL ERROR PROBABILITY p
Figure 4 E l l i p t i c a l Scale Factor vs. P r o b a b i l i t y
2.6 ConfIdence C i rc les
Let C denote a confidence c i rc le , x2 + y2 = R2, which is centered a t 0
and which has pos i t ive radius R. Then the probab i l i ty p = p(R) tha t t h e true
posi t ion f l i e s wl th in a confidence c i r c l e C i s
Def ining the aux i l ia ry parameters
K = R/ax
c = o / a Y X
-8-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELRIPSES
8 = 2 c h
y -- ( K / ~ c ) ~
and the functions
i t can be shown tha t t h i s double i n teg ra l over the c i r c l e C can be reduced t o
the s ing le d e f i n i t e i n teg ra l
a
The value o f t h i s i n teg ra l provides the so lu t ion t o questian (1) stated i n the
introduct ion.
2.7 Numerical Quadrature
Values of p(K,c) have been tabulated and are given i n the Appendix.
However, i n order t o use such a table, double i n te rpo la t i on i s required. For
values more precise than those given i n the table, the In teg ra l p(K,c) must be
evaluated numerically since i t apparently cannot be expressed i n closed
form. For d e f i n i t e i n teg ra l s o f the type p(K,c), F e t t i s [6] has shwn t ha t if
- 9 -
ALGORITHMS FOR CONFIUENCE CIRCLES AND ELLIPSES
a s u f f i c i e n t l y small step s ize i s chosen, the trapezoidal r u l e provides an
estimate f o r p(R) w i th a r b i t r a r i l y small error..
Uhenever the trapezoidal r u l e i s e f fec t i ve l y employed, Frame [7] suggests
t h a t l i nea r combinations o f the ru le w i th d i f f e r e n t step sizes w i l l provide
addi t ional estimates t o the d e f i n i t e in tegra l with only a m i n i m a l increase i n
computational e f fo r t . Such a nmer i ca l quadrature fomula i s the f i f t h -o rde r
der iva t ive corrected Simpson's ru le [lo] with step s i ze h = (b - a)/n:
2 h n
i = l + 16 1 f(a+ih-h/2)] - 60 [ f ' (b ) - f'(a)].
Since the integrand, f ( 4 ) = [eyw(+) - l ] / w ( o ) , which i s required f o r the
ca lcu lat ion of p(R), i s per iod ic w i th per iod a, i s synmetric about n, and has
continuous f i r s t der ivat ive, f ' ( 4 ) vanishes a t the end points o f the i n te rva l
o f in tegrat ion, 4 = 0 and 4 = n. Therefore, f o r the d e f i n i t e i n teg ra l under
consideration, the der iva t ive corrected Simpson's r u l e i s a l i n e a r combination
of trapezoidal sums w i th step sizes h = n /n and h = n/(2n).
The so lut ion t o question (1) stated i n the in t roduc t ion may now be
obtained by applying the trapezoidal r u l e w i t h step s i z e n/(2n) t o p(K,c),
construct ing from appropriate trapezoidal suns the der iva t ive corrected
Simpson's value, and then using the absolute value o f the d i f fe rence
- 10-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
t o estimate the maximum absolute error for the former calculat ion n h k h i s
taken as an approximation t o the p r o b a b i l i l t y p(R).
The technique o f employing the f i f t h -o rde r der iva t ive corrected Sinpson's
r u l e i n order t o estimate the error, E, i n the nunerical quadrature i s
believed t o be new and i s more e f f i c i e n t than the customary repeated ha lv ing
the step size u n t i l a s u f f i c i e n t l y small d i f ference i s obtained, since the
t r a d i t i o n a l technique uses a quadrature formula o f the same order t o
approximate the error.
t i m e i s more c r i t i c a l than on la rger computer systems.
This technique Is well suited t o microcomputers where
The required inputs f o r the ca lcu la t ion o f t he p robab i l i t y p(R) a re ul,
u2 a a , pI2 R, and n. The value o f n i s chosen so that the e r ro r estimate E
i s s u f f i c i e n t l y small. I n most p rac t i ca l appl icat ions (i.e., K - 4 and e - >
0 - l ) , a value of n = 20 w i l l resu l t i n a t leas t seven d i g i t accuracy f o r p(R).
Question (2) stated i n the in t roduct ion may naw be solved by i t e r a t i n g
I n pract ice, on the radius R(p) u n t i l the desired p robab i l i t y i s obtained.
t he i t e r a t i o n i s ac tua l l y on the aux i l i a ry parameter K - R/ux. For values o f
p(R) less than 0.9999999, K assumes values between zero and 5.7.
These considerations provide an ou t l i ne o f the theore t ica l foundation f o r
the two algorithms given i n the next sect ion f o r the ca lcu la t ion o f p(R) and
R(p) associated with confidence c i rc les.
. The ca lcu lat ions required f o r the parameters o f a confidence e l l i p s e are
straightforward and have been given i n Sections 2.4 and 2.5.
3.0 ALGOR I THMS
The f i r s t a lgor i thm solves question (1) and i s a lso referenced by the
second algorithm. The ca lcu la t ion o f 8 i n step two i s an opt ional ca lcu la t ion
-11-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
since i t i s not required f o r the computation of t he p robab i l i t y p(R)
associated with a confidence c i rc le . The input parameters are ul, u2, a, p l P B
and R.
The second a l g o r i t h solves question (2) and i s based on the secant
method. Note t h a t the i t e r a t i o n i s ac tua l l y pe r fo rmd on K which i s re la ted
t o the radius o f a confidence c i r c l e by K = R/u,. The input parameters are
01 # 42 # a s 012 s and P.
3.1 Algorithm 1 f o r p(R)
2 1. 5 = ulsin(2a) + 2plZala2sin(a)
2 2 9 = ulcos(h) + 2pl2ala2cos(a) + u2
2 % = % i n (a)
8 arctan(al/a2) (Note: use arctan (y,x) or P-R function)
-12-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES /
3. K = R / a ,
5. Select a posit ive fnteger n (e.g., n - 4)
w(+) = (c2 - l)cos($t) - (c2 + 1)
TI = f(0) + f(n)
n-1 T2 = 1- f ( i h )
1 =1
7. E = [[TI + 2(T2 - T3)]c/nI2/6
8. I f E i s suff ic ient ly small (e.g., E < loo5), accept p = p(R).
Otherwise, select a larger value f o r n and repeat steps 6 through 8.
-13-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
3.2 Algorithm 2 f o r R(p)
1.. Perform steps 1 and 2 o f algorithm 1.
2. Set 1 = 1 and select appropriate s t a r t i n g
values for the secant method (e.g. KO = 0.1, K, = 3.9. go = 0.08 - p,
and g, = 1.0).
3. Using the value o f
Ki - Ki-l Ki+l = Ki - g i g5 - giO1
where g1 = p i - p f o r 1 > 1, perform steps 4 through 8 o f algor i thm 1
t o obtain probability pit1.
4. If gi* i s s u f f i c i e n t l y small (e.g., lg i t l l < IO-'), s e t
R = R(p) =uxKi% and stop. Otherwise, repeat steps 3 and 4 w i t h
i replaced by I + 1.
4.0 NUMERICAL EXAWLES
The fo l lowing examples i l l u s t r a t e the app l ica t ion o f the two algorithms
presented i n the previous section.
4.1 Example 1
A navigator reports the ship's pos i t ion a t 41'46' N and 50'14' W.
.Assuming the angle o f crossing between the two LOPS i s a = 30'. there are no
systematic errors, and the random er rors i n the two nonorthogonal d i rec t ions
are normally and independently d i s t r i bu ted w i t h standard dev iat ions a, = 2 nm
andu2 = 1 nm, calculate the parameters o f the e r r o r e l l i p s e and the r a d i i o f
-14-
ALGORITHMS FOR CONFIUENCE CIRCLES AND ELLIPSES
t he confidence c i r c l e s f o r the various p robab i l i t i es ind icated i n Table 1.
Also compute the sizes, areas, and the probab i l i t i es associated wi th the ldRMS
and 2dRMS c i r c l e s where ldRMS = ox + or and 2dRMS = 2*1dRUS. The term ldRMS
I s also ca l led rad ia l er ror or root mean square error.
2 2 2
Use algorlttm 1 t o calculate the parameters of the error ellipse: ox =
4.3778 nm, uY - 0.9137 nm, and e = 24.5533'. The resu l t i ng er ro r e l l l pse i s
shown i n Figure 5. . .
Continuing with algor i thm 1, set n - 7 and compute ZdRMS = 11.9443 nm,
p(2dRHS) = 0.9579, and E = 4.3*10-' where E i s an estimate of t he maximum
absolute er ror i n p(R). S im i la r l y ca lcu late the values fo r the ldRMS c i r c l e
as indicated i n Table 1.
9 24.55
WRYS 4.47 D4RYS 8.94
a,& .2l
Y
X
0
Figure 3 The Error E l l i p s e
-15 -
ALGORITHMS FOR CONFIDENCE CiRCLES AND ELLIPSES
Next, use algorithm 2 t o ca lcu late the remaining values l i s t e d I n Table
1. For reference, the ldRMS and 2dRMS values computed from a l g o r i t b 1 are
included i n Table 1. See Figure 6 f o r a p l o t o f t he radius o f the confidence
c i r c l e as a nonlinear funct ion o f probabi l i ty . Note tha t 2dRMS i s an upper
bound f o r the radius o f the 95% c i rc le . The c i r c u l a r probable e r ro r o r
c i r c u l a r e r ro r probable, CEP, i s the radius o f t he 50% c i rc le .
Use the e l l i p t i c a l scale fac to r k = 2.4477 t o ca lcu late the semimajor and
semiminor axes o f the 95% e l l pse, 10.7158 nm and 2.2365 nm, respectively.
The area o f the 95% e l l i p s e i s 75.3 nm . Since the radius o f the 952 c i r c l e
i s 8.6302 nm, the area o f the 95% c i r c l e i s 234.0 nm? Thus the area o f t he
2
95% c i r c l e i s 211% l a rger than the area o f t he 95% e l l i p s e and ye t both
provide the same confidence f o r pos i t i on location.
Table 1 Parameters assoc ia ted w i th u1 8 2, a2 8 1, a = 30°, and p,, = 0
Probabi 1 i t y P
-01 .lo a50 .68218
.75 a90 .95 .95786
e99 a999 e9999 a99999
Rad1 us R
0 . 2846 0.9565 3.1033 4.4721*
5.1216 7.2604 8.6302 8.944 3**
11.3144 14.4349 17.0573 19.3592
Area A
0.3 2.9 30.3 62.8
165.6 234 . 0 251.3
402.2 654.6 914.1 1177.4
82.4
n
1 1 3 4
5 6 7 7
8 9 10 10
Error Bound E
1.3E-13 1 9E-7 6.2E-8 1.8E-7
i.a~-8 3.4E-7 2.5E-7 4.3E-7
6.1E-7 4.OE-7 1.OE-7 1 .1E-7
* l&MS ** 2cRMS
-16-
ALtiOKITHPS FOR CONFIOENCE CIRCLES AND ELLIPSES
Figure 6 Confidence C i rc les w i t h a1 = 2 , o2 = 1, a = 30°, and p12 = 0
- 1 7 -
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
4.2 Example 2
Consider a pos i t ion loca t ion system where u1 = u2 = 1 u n i t and the angle
o f crossings var ies between 0 and n. For various values o f Q and assuming
= 0, Table 2 gives the parameters o f t he e r ro r e l l i p s e and the sizes, 912
areas, and p robab i l i t i es associated wi th the 95% and 2cRMS confidence
c i rc les. Table 3 gives the areas o f these 95% c i r c l e s and e l l i pses as a
funct ion of a. Figure 7 shows a p l o t o f the radius o f the 95% c i r c l e as a
funct ion o f a . 4.3 Example 3
(See Bowditch [e].) Assuming ul = 15 m, u2 = 20 m, a = SOo, and p12 = 0,
determine the p robab i l i t y o f loca t ion w i t h i n a c i r c l e o f radius R = 30 m.
Set n = 2 i n algor i thm 1 and obta in ox = 29.8895 m, uy = 13.1023 m, 8 =
15.7733', and p(30 m) = 0.6175. The e r r o r estimate i s E = 1.3*10°7 whi le the
actual er ror i s 1.0*10 . -8
Set n = 3 i n algor i thm 2 and compute the radius o f the.95% c i r c l e : R =
60.2437 m wi th E = 1.4*106. Also, using n = 5, the radius o f the 99.9%
c i r c l e i s found t o be R = 99.3274 m w i t h E = 8.1*10°9.
The parameters o f the 95% e l l i p s e a r e ko, = 73.1620 m, kuy = 32.0712 m, 2 and0 = 15.7733'. The area o f the 95% c i r c l e , 11401.8 m i s 55% greater than
the area o f the 95% confidence e l l i pse , 7371.4 m . 2
-18 -
Table 2 Parameters associated with a1 - a2 - 1 and pl i - 0
Q X e
0.05 (-.05) 810.2848 0.5 (-.50) 81.0295. 2.5 (-2.5) 16.2108 5 (-5) 8.1131
10 (-10) 4.0721 15 (-15) 2.7321 20 (-20) 2.0674 25 (-25) 1.6732
a Radius
a c-ay/ax 2dRMS Y
.70711 .00087 1620.5702 -70713 .0087 162.0652 .rot78 ,0417 32.4526 m7098 m0875 16.2883
.7180 .1763 8.2698
.7321 .2679 5.6569
.7525 -3640 4.4003
.7802 A663 3.6922
0.1 (179.9 1 (179) 5 (175) 10 (170)
, i Prob Radius Area ' p(R-2dRWS) R(p-0.95) R(p-0.95)
-95450 1588.1292 7923 581.2 .95451 158.8165 79 239.4 .95465 3 1.7805 3 173.0
796.0 095511 15.9174
.95693 8 a140 201.8
.95986 5.4069 91.8
.96375 4.1280 53.5 -96833 3.3867 36.0
26.9 -97316 2.9266 67753 2.6458 22.0 .98059 2.4950 19.6 .98168 2.4477 18.8
20 (160) 30 (150) 40 (140) 50 (130)
2mHS 9-
1.0199 1.0205 1.0211 1,0233
1.0319 1.0462 1.0660 1.0902
1.1160 1.1376 1.1511 1.1555
60 (120) 70 (110)
90 80 (100)
i! iti%i 1 1.41421 .E165 1 S774 -1 3.2660 1.2328 ,0632 -7002 3.0099
2.8721 1.1001 .9231 '-8391 45 1.oOoo 1.0000 1~000 2.8284
8250 600.6 82 514.3 3 308.6 833.5
214.9 100.5 60.8 42.8
33.5 28.5 25.9 25.1
I I 1
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
~~
0.1 (179.9) 1 (179) 5 (175) 10 (170)
Table 3 Areas o f 95% Confidence Circles and Ell ipses when aL = a2 = 1 and pI2 = 0
810.2848 81.0295 16.2108 8.1131
a
20 (160) 30 (150) 40 (140) 50 (130)
4.0721 2.7321 2.0674 1.6732
201.8 91.8 53 m 5 36.0
55 .O 37.6 29.3 24.6
0 \=Area o f $=Area of 95% Circ le 95% El l ipse
Y
60 (120)
;; I;;:; 90
734.71 1 078.5 73.47 216.0 14 . 69 108.4
le4142 1.2328 1m1001 1 .oooo
m7180 .7321 .7525 .7802
26e9 22 -0 19.6 10.8
2 1 e 7 20 a 0 19.1 18.8
~ 3.67
1.83 1.47
I 2.44
e8165 .86 3 2 .9231
1 mO000
~ 1 e24 1 a10 1.02 1 m o o
-20-
ALGORITHMS FOR CONFIDENCE XIRCLES AND ERLIPSES
Ffgure 7 Radius of the 95% Confidence C f r c l e when a1 02 and p12 ' -21-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
4.4 Example 4
Repeat Example 3 w i t h cor re la t ion p12 = 0.5. A lgor i thn 1 with n = 3 gives uX =
36.1325 m,ay = 9.3864 m, 8 = 19.5924', and p(30 m) - 0.5666 with E = 1.9*10-8.
E = 5.8*10". The radius o f the 99.9% c i r c l e i s calculated wi th n = 7 t o be
R = 119.2786 m where E = 6.1*10-'.
Algorithm 2 with n = 6 gives f o r . t h e 95% c i rc le , R = 71.4658 m and
The 95% e l l i p s e has parameters kux = 88.4433 m, by = 22.9756 m, and 2
7he area o f the 95% c i r c l e , 16045.2 m i s 151% greater than the area 8 = 19.5924'.
o f the 95% confidence e l l ipse, 6383.8 m . 2
Comparing the resu l ts o f Examples 3 and 4, i t may be observed t h a t the e f f e c t
o f changing the cor re la t ion from zero t o 0.5 i s t o increase by 41% the area o f the
95% c i r c l e while the area of the 95% e l l i p s e i s decreased by 13%. Moreover, the
or ien ta t ion o f the 95% e l l i p s e i s increased from 15.7733' t o 19.5924'.
These examples suggest t ha t confidence e l l ipses are super ior t o confidence
c i r c l e s since they provide the same p r o b a b i l i t y o f loca t ion bu t over a s i g n i f i c a n t l y
smaller area. To be more precise, f o r any leg i t imate values o f ol, 02, a, and p12 ,
the area o f the 95% e l l i p s e i s 11*ln(400)a#~ whi le the area o f the 95% c i r c l e i s
less than the area o f the 2dRMS c i r c le , 4n(ux t uy). 2 2
I n the best o f circumstances, that i s when o1 = 02, a = n/2, and p12 = 0, t he
area o f the 95% c i r c l e i s equal t o the area o f the 95% el l ipse. However, as Example
1 shows, i n less than ideal condi t ions the 95% c i r c l e can be several hundred percent
la rger than the 95% el l ipse. Clearly, i n such si tuat ions, for any p robab i l i t y the
confidence e l l i p s e i s t o be prefer red over the confidence c i r c l e s ince a
substant ia l ly smaller area provides the same probab i l i t y o f location.
- 2 2 -
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
5.0 EQUIVALENT FORMJLAS FOR THE ERROR ELLIPSE
Def i n i ng 2 2 A o1 + 2Pl2o1o2co~(a) + o2
2 C = ol cot(a) + pl2olo2csc(a)
i t can be shown t h a t the semimajor and semiminor axes of the e r ro r e l l i p s e may
be calculated from
A c s z (a)[ A + [A2 - B2I2] 2 -
O x - 1
a y 'T CSC' (a)[A - [A2 - B2I2 ]
or
oy 'T A*cs?(a) - C*csc(a).
5.1 Special Cases
For the special case Q1 = a2 = o , and p12 = 0, It can be shown that the
parameters o f the e r ro r e l l i p s e s imp l i f y t o the fo l lawing:
-23-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
Burt, Kaplan, and Keenly's [4] and Bowditch's [2] f o m l a s f o r t h i s
special case must be used with caution since t h e i r formulas f o r ox and oy
i m p l i c i t l y require that the crossing angle between the two LOPS nust be
acute. Their formulas give incorrect resul ts f o r obtuse crossing angles.
Now i f u l = u p = u , p12 = 0, and a i s rest r ic ted t o values s t r i c t l y
between 0 and n/2, then ox and uy may be fur ther s impl i f ied t o
'L 2
a 2 O * C S C ( O / ~ )
3- u = 2 u*sec(a/2) Y
Final ly, i f u1 = u2 = u, p12 = 0, and a - n/2, then a l l calculat ions can
be greatly s impl i f ied t o the c i r cu la r normal distr ibut ion:
= a
= U
Q X
Y U
e = o
h R(p) = a [ -2* ln( l - p)] .
- 2 4 -
ALGORITHHS FOR CONFIDENCE CIRCLES AND ELLIPSES
6.0 APPLICATION TO LORAN-C
Bregstone [3], Col l ins [SI, Pierce, kKenzie, and Woo&ard [8], and
Worrell [12] state e x p l i c i t l y o r assume i m p l i c i t l y t h a t assunptions (11, (2),
and (3) l i s t e d i n Section 2.2 with p12 = 0 may be applied t o LORAN-C. Swanson
[9] also accepts the three assumptions but suggests a value o f 0.5 fo r the
co r re la t i on o f the time-difference o r TD errors.
Amos and Feldman [l] po in t out t h a t t he TD e r r o r i s a funct ion o f .many
variables. In rea l i t y , because o f t h e current design o f many LORAN-C
receivers, the centra l l i m i t theorem o f p robab i l i t y theory appl ies and i t i s
reasonable t o assume tha t the TD er ro rs are approximately n o m l l y
d is t r ibuted.
The value f o r the cor re la t ion p12 i s often taken as zero; hcmever, it i s
l i k e l y tha t another value such as 0.5 should be used. S ign i f i can t di f ferences
i n the sizes and or ientat ions o f confidence e l l i pses as we l l as t h e s izes o f
confidence c i r c les may be observed i f the cor re la t ion i s taken as 0.5 instead
o f zero.
The U.S. Coast Guard pe r iod i ca l l y publishes revised speci f icat fons o f the
I n t h i s respect, see reference Cll]. The current t ransmit ted LORAN-C signal.
value given f o r the standard dev iat ion of. the TD erors i s 108 nanoseconds.
7 m0 CONCLUSIONS AND RECOMMENDATIONS
Algorithms with new stopping c r i t e r i a have been given which may be used
t o solve two standard problems i n pos i t i on locat ion: (1) Find the p r o b a b i l i t y
p t h a t the t rue pos i t ion T i s w i t h i n a c i r c l e o f radius R centered a t the
observed pos i t ion 0; and, (2) Find the radius R o f the c i r c l e C centered a t 0
such that the p robab i l i t y i s p t h a t T l i e s w i th in C.
-25-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
It i s assumed tha t the errors associated w i t h t h e l i n e s of pos i t ion my
be approximated by a nonorthogonal b i va r ia te dependent Gaussian d i s t r i bution
where the e r ro rs are measured orthogonally t o the LOPs. The algorithms
presented f o r t h i s model are readi ly implemented on a microcomputer.
Moreover, they are p rac t i ca l since they avoid t h e use o f p robab i l i t y curves,
tables, charts, nomograms, f i c t i t i o u s functions and angles of cut, special
rat ios, sigma s t a r factors, double Langrangian interpolat ion, and Bessel
funct ions which are required by some methods o f solution.
Numerical resu l t s confirm the h igh accuracy and e f f i c i ency of t h e
algorithms presented herein f o r t he ca lcu lat ion o f t he parameters associated
w i th the e r r o r e l l i p s e and confidence c i rc les.
Confidence c i r c l e s are conceptually easi l y understood and frequently
used; however, w i t h the advent o f mlcroconputers w i t h powerful graphics
capabi l i t ies , confidence e l l i pses should be considered as a superior
a l t e rna t i ve I n appl icat ions where confidence c i r c l e s have t r a d i t i o n a l l y been
used since nuch less computation is required f o r t h e parameters o f a
confidence e l l i p s e than f o r a confidence c i r c le . kreover , t h e area of a
confidence e l l i p s e i s generally substant ia l ly l ess .than the area o f a
confidence c i r c l e having the same associated probabi l i ty ; th is can be
important not only i n rout ine pos i t ion location, but even more so, i n c r i t i ca l ’
search and rescue missions.
F ina l ly , as previously stated, the algorithms are appropriate only when
the error model described i n Section 2.2 i s v a l i d f o r the p a r t i c u l a r pos i t ion
locat ion system under consideration. Also note t h a t the algorithms nust be
modified i n s i t ua t i ons such as the m iss i l e o r t a rge t problem where the errors
are measured p a r a l l e l t o t he axes o f a coordinate system rather than
orthogonally t o the LOPs as i s t h e case i n pos i t i on locat ion calculat ions.
-26-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELL1 PSES
0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 I D 1 1.8 I .9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 3.1
,15149376 .51607270
.63187975 ,68268949 .72866788
,80639903 .83848668 .8668560 .89010142 a91086907 .92813936 .94256688
,964271 16 .97219310 .97855178 .98360193
e57628920
. m w m 6
.9n49974
.98758m7
.99m7162
.99~m605
.99488974
.99626837
.99730020
.99805479
. w i m 6
.5115181 ,572595669 .62887213 .68023254 .7266 9 6 7 .76822148 .80505480 .83740(89 ,86551266 .88970083
.92769639
.94221819
.95422722
.96405916
.97203030
.91842751
.98351079
.98750994
.99062493
.993027 12
.99486123 ,99624767 ,99728531 .9980W17
.9imioi9
.42556056 A9606835 .5644571
,67235867 ,72026823 .7630349
,834 00178 A6271282 .8875a02 ,90856194 .92631248 -9411 2996 .95337750 ,963401 12 .97152372 .97804079 .98321798 .98129005 .99016116
.99477268
.99618340
.99723907
.99802 11 9
.6igi3ni
.8008553s
.9929mi9
,38463141 .3357)0 A6332568 .417W629 .5349)877 .49418829 .59931400 .%Sly)) .6%02424 ,62912491
.75321755 .73585BO
.82770177 .81698517
.OS173618 .M930)16
.88349137 .87666446 ,90531663 m90017456 .92379894 .91972753 .e3915857 .91s8555 .95184149 .949381# .96221269 .9601709 .97061093 .9691171 .97734503 .97624187 ,98269178 .9818yW1 .98689528 .9862720( .99016742 .9897W46 .99268943 .992W833 .99461409 .99436185 .99606837 .99588118 .99715634 .997UZ662 ,99796223 .99786985
e70796818 e68599173
-79299679 .77935506
.29146823 ,25481781
.44742080 ,4021292
.52139984 .I7593757
.59009533 ,51613196
.65244889 .61163161
.70799732 ,67142689
.7%6)26656 .724%735
.79892881 J7208895
.86575592 ..e4783930 a89155362 a87131164 .91306800 .901911@ .930B6154 .92222712
.36993053 a328WW2
. ~ ~ s m i 6 0 .ai292873
.94nn1 .wsm
.95732052 .95229986
.96688448 .96310169
.97452393 .97169345
.98057026 .97846612
.98531115 ,98375690 ,98899336 .98785268 .e9182603 .994)9441
-99561263 .99517918 .99682936 .99652052
.9939~23 .W~)BPOQ
.99772961 .99751096
.22511143
.29256543
.M27l224 m43336291 .50257901 .SB74674 ,63061681 .0731223
.7833%28
.I2262457 ’ &624112
.E8466237
.908%088
.92787988
.94376684
.OS655220
.97474955
.98100352
.73e3m94
.96673m3
.2oQ)7981 e18117832
.26293740 .2#)15834 .e32834532 ,29897012 -39532797 e36201358 .U214212 A2575533 .52124621 A8878140 ,58934913 .54987365 ,64743948 .60798223 .70078999 .66230358 e74895002 m11225465 ,79171937 .75147088 ,82911370 ,79778816 ,86132384 .83121750 .MB67314 .E6391495 .91157619 .89014951
.94593857 .930682ll ,9305133 .gin7137
,958373ae .9458mu .9~126981 .gsamo39 e97605221 e96791351
.98583311 .98210228 .97#9685
.98952681 .98675296 .98178371
.99232491 A9020880 .98648759
.W72919
.2I729546
.21150%
.)uop319
.39w6934
.45392557 .51l4774 .nww .6246M90 .67uI7U .721%270 a16425392 ALPIOI~O
. W 7 2
.m749(7
.911m83e
.92899465
.94m6524
.9%a6307
.%595255
.97387859
.99442459 .99294021 .9900816 .98507921
.99713480 .99638509 .99481678 .99181130
.83552551
. t s m s g i
.ooswni . 9 9 4 9 2 i ~ .992792s . 9 m i o o
3.6 ,99968118 3.7 .99978440 3.8 .99985530 3.9 .99990381 4.0 .99993666 4.1 .99995868 4.2 .99997331 4.3 .99998292 4.4 ,99998911 4.5 .99999320 4.6 .99999578 4.7 ,99999740 4.8 .99999841
5.0 .99999943 4.9 .99wggm
5.1 .99999966 5.2 .99999980 5.3 .99999988 5.4 ,99999993 5.5 .99999996 5.6 .99999998 5.7 .99999999 5.8 .99999999 5.9 1.oooooO0O 6.0 6.1 6.2
.99968007
.99918324 ,99985453 .9999a)29 .99993632 .99995847 .99997317 .99998283 .99998912 .99999317 ,9999957 5 .9999978 .99999841
.99999932
.999999m
.99967476 .99966527
.99971965 .99977325 ,99985213 .99984785 ,99990170 .99989886 .99993527 .99993341 .99995179 .99995657 .99997273 .99997195 .99998255 -99998205 .99998894 -99998863
.99999568 .99999556
.99999lW .99999727
.99999838 .99999833
.99999902 .99999899 -99999941 -99999940
.999993m .99999286
,99999966 .99999965 ,99999980 .99999980 .99999988 .99999988 .99999993 .99999993 .99999996 .99999996 ,99999998 .99999998 .99999999 .99999999 .99999999 .99999999
l.Woooo00 1.oooooooo
-99999964 ,99999979 .99999988 .99999993 .99999996 .99999998 .99999999 ,99999999
1.ooooww
.99%5017 .99962813 .999763ZS .99974820 .99981117 .99983111 . 9 9 9 8 W .99988778 .99993Sl -99992614 39995468 .99995185 .99997073 .99996890 -99998127 .99998011 .99998B13 .99998740 .99999255 -99999209 .99999537 .99999508 -99999715 .99999697 -99999826 -99999816 .9999095 .99999889 .99999937 .99999933 -99999963 -99999961 .99999970 .99999977 .99999987 .99999987 -99999993 -99999992 -99999996 -99999996 .99999998 .99999998 .99999999 .99999999 39999999 .99999999
1.0000004) 1.00000000
,99959377 .99953644
.99981%8 .99979017
.99987758 ,99986018
.99991946 .9999C%49
.99994751 A9994011 -99996611 -99996156 ,99997833 .99997544 39998628 .99998445 ,99999139 .99999025 .99999465 .99999395 .99999611 .9999%28 .99999199 .99999773 .99999879 .MW3 .99999928 -99999918
.99972508 . w 9 6 e m 39942181 .99961019 .99913%0 .99982765 .99988697 .999926% .99995273 .999%985 .99998095 .99998808 .99999261 -9999996 -99999724 .99999833 .99999901
.99914419
.999420&(
.99961195 -99974257 .99983090 .99989oop -99992917 A9995483 A9997147 -99998216 -99998095 -99999322 .99999588 .99999752 .99999852
. 9 9 w 1 9
.99893523
.99926€30
. W Y
.sses5225
.99990)11
.99993748
.9999%93
.99997458
.999mm
. M o o )
.9999999
.99999621
.wwa
.oosnus
,99999957 .99999952 -99999941 .99999913 . M 7 7 5 .99999975 .99999972 .99999966 .99999949 .99999CM ,99999985 .99999W -99999980 ,99999971 .99999921’ .99999992 .99999991 -99999989 .99999903 .99999953 .99999995 .99999995 .99999993 .99999990 .99999973 .99999997 ,99999997 .99999996 .99999995 .99999985 .99999998 .99999998 39999998 .99999997 .99999991 .99999999 ,99999999 .99999999 ,99999998 .99999995
1.00owo000 .99999999 .99999999 .99999999 .99999997 1.00000Ooo 1.00000000 1.ooO0Oooo .99999998
.999(39999 1.-
p(K.c) = probablllty that a polnt l l e s wlthln a c l r c l r I h s e center i s at the origin and rhose radlus I s I KO,. Mere c 0 /a where a, i s the larger standard dcvlatlon. Ihc table glrw r o l u n of the standard orthogonal b l rar la te lndcpmdent Gausslan distrlbutlon.
Y ‘
-27-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
REFERENCES
1. Amos, 0. and Feldman, 0. * A Systematic Method o f LORAN-C Accuracy Contour Estimation." Washington, D.C.: U.S. Coast Guard, unpublished report, pp 1-18.
2. Bowditch. Nathanlel Ameiican Pract ica l Navigator. Washington, D.C.: Defense Mapping Agency Hydrographic Center, Vol. 1, (1977) pp 1204-1237.
3.
4.
5.
6.
7.
8.
9.
10.
11 .
Bregstone, Edward Washington, D.C.: U.S. Coast Guard, telephone comn ica t i on , (1981) .
Burt, W. A., Kaplan, 0. J . , and Keenly, R,* R., e t a1 "Mathematical Considerations Per ta in ing t o the Accuracy of Pos i t ion l oca t i on and Navigation Systems.' k n l o Park: Stanford Research Ins t i t u te , Naval Warfare Research Center Research Memorpdum NWRC-RM 34, (1965) pp 1-21.
Col 1 ins, James "Formulas f o r Posi t ion ing a t Sea by Circular, Hyperbolic, and Astronomic Methods." Washington, D.C.: National Oceanic and Atmospheric Administration,NOAA Technical Report NOS81, (1980) pp 1-26.
Fet t i s , Henry E. "Numerical Calculat ion o f Cer ta in D e f i n i t e In tegra ls by Poisson's Summation Formula.'' Math. Tables and Other Aids t o Computation, 9 (1955) pp 85-92.
Frame, 3. Sutherland ' "Numerical In tegrat ion and the Euler-Maclaurin Summation Formula." East Lansing: Michigan State University, unpublished report, pp 1-9.
Pierce, 3. A., McKenzie, A. A., and Woodward, R. H. - LORAN. M I T Radiation laboratory Series Vol. 4, New York: McGraw- H i l l Book CO., (1948) pp 425-431.
Swanson, E. R. "Estimating the Accuracy o f Navigation Systems. A S t a t i s t i c a l Method o f Determining Ci rcu lar Errors Probable." San Diego: U.S. Navy E l e c t r i c a l Labs., Research Report 1188, (1963) pp 1-5.
"An E f f i c i e n t Modi f icat ion o f Euler-Maclaurin's Formul a." Trans. Japan Society f o r C i v i l Engineers, 24 (1955) pp 1-5.
U.S. Coast Guard "Speci f icat ion o f the Transmitted LORAN-C Signal." Washington, D.C.: Department o f Transportation, U.S. Coast Guard COMDTINST M16562.4, July 14, 1981.
Tanimoto, B.
-28-
ALGORITHMS FOR CONFIDENCE CIRCLES AND ELLIPSES
12, Worrell, Clarence L. "Production of LORAN-C Reliability Diagrams a t the Defense Mapping Agency." Washington, D.C.: Defense Mappfng Agency Hydrographic Center, unpubli shed report, pp 1-9.
- 2 9 -