Statistics of A Geometric Representation
Statistics of A Geometric Representation
Statistics of A Geometric Representation
A precise statistical definition is established for the geometric "shape" of a randomly distorted wavefront.
Relationships between the phase-structure function and the statistics governing the shape are derived. The
most significant portion of wavefront distortion caused by atmospheric turbulence is a random tilting of the
plane-wave front. A procedure is outlined for calculating the influence of wavefront distortion on optical
systems. Estimates are formed of the effect of wave-front distortion on photographic resolution and optical
heterodyne efficiency.
The integration in Eq. (2.3) and all other integrations =1j (3.5)
throughout this paper are understood to cover the
infinite plane, unless explicitly indicated otherwise. where j takes the values C, L, S, and Q and nj has the
Finite bounds for the integraion are provided, in effect, values 1, 3, 4, and 6, respectively. The geometric
by the aperture function. significance of the four Ij's defined above is obvious.
Now, let us define Ai as
III. WAVEFRONT APPROXIMATION
The distorted wavefront's phase 4 (x) is approximated Aj-=(l/rR2) dxW(xD)[0(x)-cPj(X)]2. (3.6)
by the infinite series
(D(x)= E a,,F.(x), (3.1) If the ensemble average (Aj) can be computed, we obtain
A1=1 a measure of the goodness of various possible finite
where the coefficients a. are chosen to optimize the series approximations to the distorted wave front.
approximation. The error in the approximation is A, the
aperture-averaged-square difference between >(x) and IV. DERIVATION OF UNAVERAGED
RELATIONSHIPS
5 M. Born and E. Wolf, Principlesof Optics (Pergamon Press, Substituting for /j(x) in Eq. (3.3) by using Eq. (3.1),
Inc., New York, 1959). expanding the square, and interchanging the order of
November 1965 STATISTICS OF WAVEFRONT DISTORTION 1429
summation and integration, we find From Eqs. (3.4b, c, and d) and Eq. (4.4') we can see that
2
(aL) == AC-AL, (4.6a)
dxW(x,D)q2 (x)
A=-
(as) 2 = AL- AS, (4.6b)
2 RJ (aQ) 2 = AL-AQ. (4.6c)
- aa,, dxW(x,D)O(x)F,(x)
=rR2 J The problem at this point is reduced to computation of
the ensemble average of Aj as given by Eq. (4.5') and
+ A aY a- b dxW(x,D)F,(x)F^(x). (4.1) relating the result to the phase-structure function.
$C=1 7rR 2
V. DERIVATION OF AVERAGED
Invoking the orthonormality condition, Eq. (2.3), we RELATIONSHIPS
can perform the last integration and then one of the FE~amining Eqs. (2.2a-f), we note that FI(x) is inde-
two sum~mations in the double sum, we obtain
pendent of x so that Jfdx'W(x',D)F 1 (x)F1 (x') equals
11- unity. Also for g not equal to one, JfdxW(x',D)F,(x)
'dxW(x,D)02(x) XF,(x') vanishes. Consequently,
7rR 2 J
2 .0 1 1
- 2
A a,; dxTW(x,D)Ok(x)F,,(x) - fdxW(x,D)k2(x)= ffdxdx'W(x,D)
,7rR2 '_1 7rR2 J rR 2 J
1 0
+ 2
E_ (a,,E2~. (4.2) XTV(x',D)Y F,,.(x)F,,(X'),02(X). (5.1)
7rR g=1 i1-
Substituting Eq. (4.2) into Eq. (3.2), differentiating, Symmetrizing between x and x' in the right-hand side
and solving for a,, we get of Eq. (5.1) gives
Working with the finite series approximation, we can XE F,l(r'+ r)F,,,(r'- 1r) O(r) . (5.4)
8t=1
easily show the results corresponding to Eq. (4.4) and
(4.5) to be The r' integration in Eq. (5.4) can be performed ex-
plicitly. For this purpose, define the function 5#(r,D) as
1 1- 1 nlj
A= IdxW(x,D)0 2 (x)- Y(a.)2 (4.4')
7rR2 J 7rR 2
g=1
5#(r,D)= fdr'W( Ir'+±rf,D)W( r'-1rj,D)
A=-
1 1.1C
dxW(x,D)02(x)-- J dxdx'W(x,D) nX
XA F,(r'Jr r)F,,(r'- Ir). (5.5)
pull
S, and Q. The results are ignore this factor. (A discussion of this matter is given
1 in Appendix D.)
5Fc(r,D)=(1/7r){2 cos- (r/D) In this paper, particular values of -A are not of con-
-2(r/D)[1-(r/D) 2 ]i} W(r,2D), (5.6a) cern. (The necessary data for computing A are provided
9L(r,D) = (1/7r){6 cos-'(r/D)- [14(r/D)-8(r/D)3 ] in Appendix C.) In fact, it is convenient to replace -A
with a new quantity r0 which has the dimensions of
X E1- (r/D)2]} W(r,2D), (5.6b) length and is defined by
gs(r,D)= (1/7r) (8 cos-1 (r1D) rOc-(6.88/eA)3 5. (6.2)
-E24(r/D)-(80/3)(r/D)3 +(32/3)(r/LD) 5 ]
X[1-(r/D)2 ]1 }2W(r,2D), (5.6c) Correspondingly, the phase structure function, ex-
pressed in terms of r0, is
9Q(r,D)= (l)1 2 cos-1 (r1D)
D(r) = 6.88(r/ro)5 13 . (6.3)
-E44(r/D)-64(r/D)+±32(r/D) 5]
X 1-(r/D)2 ]} W(r,2D). (5.6d) The apparently arbitrary constant 6.88 was chosen on
the basis of the analysis of the performance of an optical
The evaluation of 5:j(r,D) given in Eqs. (5.6a-d), ob- heterodyne detection system.' In Ref. 4, ro is shown to
tained in the Appendix through a rather extensive and be that diameter of a heterodyne collector for which
quite uninteresting calculation, is fundamental to the distortion effects begin to seriously limit performance.
program of this paper. With this in hand, the rest of the It is seen later that it is also that diameter for which
work is straightforward. Substituting from Eq. (5.5) (Ac) is essentially unity. Typical values of ro for visible
into Eq. (5.4) and noting that the integrand is isotropic and near infrared wavelengths and for approximately
in r, so that the angular integration can be performed, vertical propagation paths down through the atmos-
we find phere are of the order of several centimeters.
(Aj) =
1
f
rD
rdrgj(r,D)D(r), (5.7) VII. EVALUATIONS
Substituting Eq. (6.6) into (5.7), replacing the
1 D
variable r by it, where
((aL )2) = - rdr[9c(r,D)- 5L(r,D)]5t(2 ), (5.8a)
u= r/D, (7.1)
rD
and noting that
(
((as)2) =- rdr[5L(r,D)-5is(r,D)]°)(r), (5-8b) 3j(r,D) = 5j(ul), (7.2)
we get
1 rD (Aj)=27.5(D/ro)513 udu~j(u,1)u5 /3 . (7.3)
((aQ)2) = R J rdr[IEL(r,D)- 5:Q(r,D)]D(r). (5.8c)
spatial separation, it is possible to show that the phase where B(a,O) is the well-known beta function, and
structure function may be written as defining Ij as
5)(r) =,_*r5/3 (6.1) _Tj = it"15j(ul)du, (7.5)
whereXA is a constant determined by the path of propa-
gation, the wavelength, and the particular environ- we find that
mental conditions. Equation (6.1) is exact only in the Ic!3.68X 10-2, (7.6a)
near field; i.e., for short propagation paths. For longer IL-4. 73
X 10-3, (7.6b)
distances of propagation, there is an additional r-de-
pendent factor which varies from one half to one, which IS-3.96X 10-3, (7.6c)
is suppressed in (6.1). For the body of this paper, we IQc2.29X 10-. (7.6d)
61A. Kolomogoroff in Turbulence, Classic Papers on Statistical
Thzeory, edited by S. K. Friedlander and L. Topper (Interscience 7 W. Grobner and N. Hofreiter, Integraltafel (Springer-Verlag,
Publishers, Inc., New York, 1961), p. 151. Berlin/Vienna, 1961), Vol II, Eqs. (121.1) and (341.5a).
November 1965 STATISTICS OF WAVEFRONT DISTORTION 1431
mean square phase deviation used to compute the computed from the phase-structure function. Formulas
Strehl definition should be taken as (AL), not (Ac). have been provided for these computations for the first
From Eq. (7.11b) it is seen that 30% Strehl definition four types of deformation. It has been shown that
for very short exposures is achieved when the lens according to current theory of optical propagation in the
diameter is about 3. 40ro. The corresponding angular atmosphere a large part of the wavefront deformation
resolution is of the order of X/3.4ro, which is 3.4 times can be understood as wavefront tilting. Sample tech-
as much as the long-exposure resolution. (In a more niques for interpreting the deformation results in terms
precise calculation considering the effect of wavefront of performance of certain types of optical systems have
distortion on the integrated modulation-transfer func- been provided, the implication being that quantitative
tion, we have found that the resolution for high- estimate can be generated of the performance of any
speed exposure is increased by a factor of 2, not 3.4, type of optical system whose behavior is limited by
over the best that can be achieved with a long exposure. wavefront distortion.
The peak resolution is achieved at a lens diameter of
approximately 3.8ro.) ACKNOWLEDGMENTS
For an optical heterodyne receiver, as indicated in
I would like to take this opportunity to thank J. D.
Sec. VI, the performance starts to deteriorate signifi-
Cloud and M. F. Sternberg for assistance in tracking
cantly when the collector diameter becomes larger than down
errors in the evaluation of 5s(r,D) and 5Q(r,D),
ro. This can be understood from Eq. (7.7a) which indi- and for computation
of the 11's.
cates that the rms phase deviation for this size aperture
is about one radian and from consideration of the fact APPENDIX A. EVALUATION OF 5:j(r)
that if the rms difference is this large then there are
many pairs of points in the aperture where the phase In this appendix we outline the procedure used in
difference is 7r. The signal generated by these pairs of evaluating the function 5:(r,D) for j= C, L, S and Q.
points is very small, contributing very little to the total As intermediate steps in the process, we have to evaluate
performance of the detector. the fuictions K#(r,D) for i=0, 1, 2, and 3; J(m,n; r,D)
For a heterodyne receiver which can track at a suffi- for (m,n) = (0,0), (2,0), (0,2), (4,0), (2,2), and (0,4); and
ciently high rate any tilting of the signal wavefront, the L(m,n; r,D) for (m,n)= (4,0), (3,0), (2,0), (1,0), (0,0)
mean square phase deviation as far as the system per- (2,2), (1,2), (0,2), and (0,4). These functions are
formance is concerned, should be computed from Eq. defined as
(7.7b) for (AL) rather than from (Ac). The rras phase
deviation, according to Eq. (7.1 lb) does not become one
Ko(r,D) = fdr'W(J r'+Ir, D)W(I r'- r, D), (Ala)
radian until the collector diameter is 3.40ro. Thus a
tracking system could utilize a collector diameter 3.4
times as large as a nontracking system. This would 2
yield a shot-noise-limited signal-to-noise-ratio improve- Ko(r,D) = fdr'W(1r'+'rI,D)W(|r'--rJ, D)r' , (Alb)
ment over the nontracking system of 20 log3.4~Ž10 dB.
(A more precise calculation would probably yield a
somewhat smaller improvement.) K 2(r,D)= dr'W(|r'+4rj,D)W(Jr'- r D)r'4, (Atc) ,
By use of a rapidly adjustable variable focal-length
lens system, spherical deformation of the collected
wavefront as well as tilt could be tracked out. However, K 3(r,D) = fdr'W( Ir+ 2r |, D)
comparison of Eqs. (7.1lb, c) shows that the resulting J XW(|r'-4rJ, D)(r.r')2 , (Ald)
gain over a system which only tracked tilt would be
rR-!r r+[R2-(R+,',)2]4
trivial.
J(nnl; r,D) = 2] dp dqpn7ql, (A2)
Similar semiquantitative analyses of the effect of
J0 J-R2-(R+A'r)'] !
wavefront deformation on other types of optical sys-
tems can be computed in the same manner as the cases
treated in this section. L(n,,n; rD)=f dzvv(l- V2 )(n+l)I 2 . (A3)
In the discussion of the imaging and the heterodyne SbD
detection systems, it has been assumed that intensity Starting from Eqs. (5.3a, b), we can show that
variations across the aperture were not present, or could
be ignored. This consideration is discussed in AppendixD. 1ic 1
E Fi(x)Fi(x') = -, 2 (A4a)
IX. CONCLUSIONS i=1 7rR
ns 1 see that
2
Z Fj(x)Fj(x')= [l2r4+±(6r2- 8R2)r
E
i=7rR6 J(0,0; r,D)=4R2[L(O,O; r,D)], (AlOa)
4
J(2,0; r,D) = 4R [L(2,0; r,D)-2(r/D)L(lO;r,,D)
+ (3r4 -4R 2r2+4R 4)-12(r-r') 2 ], (A4c)
+(r/D)2 L(O,O; rD)], (AlOb)
flQ 1
J(0,2; rD)= !RIEL(0,2; rD)], (A10c)
Z Fj(x)Fj(x')=-(18r'
6
4 +(-3r2- 8R2)r'2
i=l1 7rR
J(4,0; r,D) = 4R6 [L(4,0; r,D)- 4(r/D)L(3,0; r,D)
+ [(9/8)r4-4R r +4R 2 2 4 ]-6(r. r') 2 }). (A4d) +6(r/D) L(2,0; r,D)-4(r/D)3 L(lO; r,D)
2
Expanding [v-(r/D)]m and considering Eq. (A3) we = X2+y2+ X12+Y12 - 2xx'- 2yy'. (B1)
1434 D. L. FRIED Vol. 55
This can be expressed in terms of the functions F,(x) sequently depends on A because altitude may vary along
and F,(x') giving the path of propagation. CN2 may be written in terms
of the atmospheric refractive-index variance A and the
r 2= [ 2 R8 /12][1[l(x)F.4(X')+Fl(x')F
4 (X)] outer scale of turbulence Lo defined here as the distance
+ 7rR 4I,'1(x)F,(x') - (rR 4 /2) in which turbulence correlation falls to one-half of the
X [F2(x)F2 (x')+F3 (x)F 3 (x')]. (B2) maximum value. This expression is
We note that because of the orthonormality conditions CN2 =AL&-'. (C2)
A reasonable fit to available data for A and L0 is ob-
1
27rR 2 JJ
- dxdx'W(x,D)W(x',D)r'
n
E F,.(x)F,,(x') tained with expressions
'l = 6.7 X 10-'4 exp(-Iz/3200), (C3a)
2
R /2 if j=C L,= 2/0, (C3b)
0 if j C, where h is altitude, and all units of measure are under-
stood to be in mks units. Substituting (C3a, b) into
since the F1F4 term cannot contribute, the FF, term Eq. (C2), and the result into Eq. (Cl), we obtain a
always contributes R 2/2 to the result, and the F2F2 and closed-form result for -A. This result for the most general
F3,F terms each contribute - R2 /4 whenever nj is three type of propagation path (ignoring curvature of the
or greater; i.e., whenever j#C. In this latter case, the earth) is expressible in terms of the incomplete gamma
net contribution is zero. However, transforming to r and function of order two-thirds.
r' coordinates and considering the definition of 5F#(r,D)
as given in Eq. (5.5), we see that APPENDIX D. SCINTILLATION EFFECTS
Equation (4.1) now takes the form Offhand, while pleased with the increased precision
of the structure function for propagation in the atmos-
phere, obtained by considering scintillation and working
fA=- 2 2
dxW(x,D)[0 (x)+1 (x)] with the complex phase we have to ask if we can now
assign any meaning to the complex polynomial used to
match the complex phase. For a problem in which there
- JdxW(x,D) is no scintillation, or when the propagation path is
-7rR2-=I
short, the results derived in the main body of the paper
X {f[(x) - il(x)]d,,+ [E(x)+il(x)]a,,}F,,(x) are correct with 5O(r) interpreted as the phase-structure
function and we can assign exact physical meaning to
1 /0 the quantities we have been evaluating. For long-path
+- E a,a^ dxW(x,D)F,(x)F,(x), (D2) propagation, where scintillation may be significant, we
7rR 2v~
make no more than a factor-of-two error by still inter-
where the overbar denotes a complex conjugate. Equa- preting aD(r) as the phase-structure function, and we
tion (4.2) takes a corresponding form. Now, applying retain some of the physical insight into the quantities
the modified form of (3.2) to determine the real and evaluated.
imaginary parts of a,, separately, we find that Eq. To the extent that only phase fluctuation is of im-
(4.3) takes the form portance in a process, we overestimate the extent of the
distortion by a factor of two. In those processes in which
f
a,,= dxW(x,D)[4(x)-il (x)]F,,(x),
intensity fluctuations also are significant we may not
be overestimating the effect of distortion. In the case
of the image-forming process of a lens, for instance,
as might have been expected. All the rest of the equa- intensity variations across the lens result in additional
tions of Sec. 4 follow immediately with k(x) replaced diffraction which limits resolution. If the intensity
by [O(x)-il(x)], and squares understood to involve variations are random, the log amplitude appears in
absolute values. image statistics in a form analogous to the phase so that
Likewise, all the equations of Sec. 5 follow with the the wave-structure function is the quantity which
phase 0 (x) replaced by the complex phase [4(x) - il(x)], appears. Thus, use of Dc*, as computed from (6.1) and
and with D(r), previously the phase-structure function (Cl), to estimate photographic resolution should be
as defined in (1.1), now understood to be the wave- more precise than the derivation, based on phase
structure function as defined in (Dl). Thus Eqs. (5.7) fluctuation alone, would justify. On the other hand, the
and (5.8a-c) are seen to be correct as stated when scintilla- use of DL*, when intensity fluctuations are significant,
tion is present except that 5)(r) has a different interpreta- to estimate high-speed resolution may be more opti-
tion than given in Eq. (1.1). mistic than warranted since the high-speed exposure,
The balance of the paper is better with this inter- freezing the distortion, can compensate only for wave-
pretation of 5(r) than it was initially. Equation (6.1) front tilt, but not for a linear variation of log amplitude.
and (Cl) are precise when D(r) denotes the wave- When intensity fluctuations are not significant, i.e.,
2
structure function. If D(r) is simply the phase-structure when ([l(x)-l(x')]2)«([k(x)-4(x')] ), the use of DL*
function then these two equations are accurate only to estimate high-speed photographic resolution should
when the propagation path is short. For a long path, the be quite accurate.
coefficient a contains an r-dependent factor which varies How to use the geometric statistics of wavefront
slowly and monotonically from a value of about one-half distortion, when to ignore, and how to consider the
at r = 0 to unity for large r. This factor was arbitrarily effect of intensity variations requires careful judgment
set equal to unity in (6.1) and (Cl). by the user.