LJBflARY
TECHNICAL REPORT SCCTKXI
NAVAL POSTGRADUATE SCHOOL
MONTEREY. GAUFOOMIA S&940
NPS-61Md73111A
NAVAL POSTGRADUATE SCHOOL
//
Monterey, California
15
Nov
1973
NPS-61Md73111A
Pred
icting Sound ]
Dhai
se and Amplitude Fluctuations
due
to
Micros tructure
in
the
s Upper Ocean
Physics & Chemistry
H.
Medwin
Department
Approved for public release; distribution unlimited.
FEDDOCS
D 208.14/2:NPS-61MD73111A
NAVAL POSTGRADUATE SCHOOL
Monterey , California
Rear Admiral Mc B. Freeman M0 U. Clauser
Superintendent Provost
TITLE: Predicting Sound Phase and Amplitude Fluctuations Due to Micro-
structure in the Upper Ocean
AUTHOR: H. Medwin
ABSTRACT:
Tne temporal and spatial variations of the index of refraction cause fluctuations
of sound phase and amplitude that can be completely understood only by defining
the index in terms of the duration, location, range and time of the acoustic
experiment. A truncated "universal" spatial correlation function of the index has
been derived from a simplified form of the Kolmogorov-Batchelor spectrum of
temperature fluctuations in a homogeneous, isotropic medium. Although this
correlation function is shown to be predictable simply from the depth of the
experiment, it is of only limited validity with respect to large spatial lags.
However, a Gaussian extrapolation of the "universal" correlation function
together with the standard deviation of the index provides simple useful predic-
tions of the sound fluctuations due to temperature micros tructure in the upper
ocean.
This task was supported by Naval Ship Systems Command (Code PMS 302).
TABLE OF CONTENTS
1. Introduction: Sources of Sound Phase and Amplitude Fluctuations
1.1 Microstructure in Mixed Water
1.2 Layered Microstructure
1.3 Other Underwater Structural Features
2. Wave Number Spectrum and Spatial Correlation of the Index of
Refraction in a Mixed Layer
2 . 1 The Wave Number Spectrum
2.2 The Spatial Correlation and Structure Functions of the
Microstructure
3. Sound Phase and Amplitude Fluctuations Due to Microstructure
3 . 1 Phase Fluctuations
3.2 Amplitude Fluctuation
3.3 Small Amplitude Fluctuations: Laboratory Experiments
3.4 Large Amplitude Fluctuations
3.5 Fluctuations at Sea
3.6 The Limits of Predictions
4. Conclusion
References
Initial Distribution List
Form DD 1473
1 . Introduction
Sources of Sound Phase and Amplitude Fluctuations
As it proceeds away from its source, a sound at sea encounters changing
values of temperature, salinity and density, as well as entrained objects and
bubbles in motion. These acoustically significant quantities are constant
neither in space nor in time. Slower than the microscale fluctuations due
to surface waves and turbulence of the medium, which show perceptible changes
during a time scale of the order of seconds , are changes due to internal waves
with periodicities of the order of minutes , tidal and diurnal variations , and
seasonal changes. Depending on the duration of the study, some of these
fluctuations may appear to be simply periodic, some show the spectral char-
acteristics of a narrow band noise, many are describable only by statistical
methods , and still others, such as the appearance of schools of fish or the
incidence of storms, are largely intermittent in character. (Weston 1969).
It is important to make the point that it is not the inhomogeneities but rather
it is the temporal change in the position or character of the inhomogeneities
that is the source of interest in this report.
The phase of a sound wave travelling through a stationary medium depends
on the speed of sound along the path. This dependence is expressed by
equations 1.1 and 1.2:
^^ ds = «• ^fiw ds
(x) has been shown to be (Batchelor 195 6) uniquely defined by the two
parameters, the kinematic shear viscosity, v , and the rate of supply (or
removal) of energy, e . The equation for the one dimensional PSD is
*v(x) = bx"5/3 (2.1)
where b = b {v , e)
x = 2tt/A= wave number of a turbulent
velocity component
A = effective wave length of a turbulent
velocity component
€ = energy dissipation, per unit mass, per unit time
V = kinematic shear viscosity of water
The -5/3 rds power law has been amply verified for components of turbulent
velocity in laboratory experiments and, to a lesser extent, at sea.
We are interested in the turbulent velocity not only because it is a
significant source of sound fluctuations in special situations such as turbu-
lent channels but, more importantly, because it is a guide to the behavior of
temperature variations. And temperature variations are generally the principal
sources of sound fluctuations in the ocean.
The temperature often acts as a convected, passive contaminent which
closely follows the turbulent velocity behavior. The spectrum of temperature
fluctuations has been studied for such a case (Batchelor 1959) and again the
-5/3 law is predicted for the inertial subrange. The predictions have been
largely verified (Stewart 1962, Grant 1968) by ocean measurements.
4
o
Fig. 2 Temperature and velocity spectra at depth 2 7m in the open sea, from
Grant (1968a). The predicted transition wave number, x. /is shown,
as calculated from Eq. (2.8).
Fig. 2 plots the spectra (Grant 1968) of a component of velocity and tempera-
ture measured at 27m depth, compared with the theories described in the
references. The figure shows the extension of the Kolmogorov theory from
the inertial subrange, where the -5/3 slope is confirmed, into the higher wave
number dissipation sub-range. The boundary between the two subranges is often
assumed to be at the Kolmogorov wave number,
= H
b
= (e h3)k (2.2)
Values of € , derived from measurements of the turbulence spectrum, are
a function of depth and have been found (Grant 1968) to be as high as 0.5
cm sec atjan acean depth of 15 meters in well-mixed water and as small
as 1.7 x 10 cm in quiet water below the mixed region at 213 meter depth.
Looking again at Fig. 2 we see a rise in the temperature spectrum, also
predicted Jpy BatcheLpr (1959). At this depth, 27m, £ was determined to be
5.2 x 10 cm /sec . Assuming v = 1.4 x 10 " cm /sec, Eq. 2.2 yeilds
x =6.6 cm . The rise in the temperature spectrum occurs between x and
S _2 °
x —2.4 x 10 x . At greater depths, the dissipation constant and the Kolmogorov
wave number are generally smaller.
Having identified the high wave number end of the isotropic -5/3 spectrum,
we now seek to establish the low wave number limit. Unfortunately, at the low
wave number end of the spectrum of temperature micro structure, the PSD is
very dependent on the local conditions and history of the water mass, and only
empirical results are available to guide us.
A useful, but very crude, generalization that can provide some guidance
is that the largest temperature wave length that could conceivably be
isotropic at the depth H would be of extent 4H. The corresponding isotropic
wave number would be
2TT TT in i\
This would suggest that for Fig. 2, where H = 27m, the lower limit of the
-5/3 spectrum might be at
Hm = 5*8 x 10~ cm_1 (°r lo8io Hra =-3-2).
In fact, the data of Fig. 2 show that at values of x much smaller than 5.8 x
10~ (perhaps at x- 0.02 cm" ) both the temperature and the velocity PSD
break away from the -5/3 slope. This suggests that, realistically, the lower
10
end of the -5/3 spectrum to be expected at sea is somewhere between x and
x . We will call that transition wave number, x .
o t
The appropriate variable to describe sound phase and amplitude fluctuations
will turn out to be the speed of sound, and so it is the spectrum of fluctuations
of the speed that we must seek. The connection is direct. For simplicity we
will assume that the speed depends only on the temperature. We have previously
seen that the fluctuations in temperature micros tructure are much less than the
average temperature. The fluctuations in the speed of sound are also a very
small fraction of the average value. It therefore turns out to be most useful to
define not only the index of refraction of the sound speed, at position U
— ♦
ft(l)s _i^_ (2 < 4)
— +
but also the excess index of refraction, \i (R) , which will be the particular
parameter for all future discussions of fluctuations.
H (R), c <»>" ic <*)> = n ft. k< !
The mean value, ( ) , is to be calculated over the time of the experiment. It
is essential to realize that [± (R) is sensitive to position R* . However, for
typographic convenience we will drop the notation for the explicit dependence
on position so that henceforth we write p. (R) =|i. Further, we define
/ \ a * - I 2\z (2-6)
(p,) = o and a = \\t, )
The point must also be made that if the time or duration of our averaging
process is changed, the value of <|_i) and a will change. We must immediately
accept that these quantities must have their temporal limits restricted. A
logical criterion for the restriction is that for the present we are concerned
only with locally homogeneous regions of the ocean, only for times comparable
to the duration of the acoustic experiment. We will be more precise about
these limitations , shortly.
Since the excess index, \i, is proportional to the excess speed of sound,
which in turn is assumed to be proportional to the excess temperature, AT, a
spectrum of M- can be expected to resemble the spectrum of AT. It will have an
inertial subrange in which the log of the power spectral density, $ (x) will have
a slope of -5/3 when plotted against log x , as in Fig. 2. That isotropic inertial
subrange will have an upper limit x* and a lower limit which we will call x^.
11
The lower bound of the isotropic inertial subrange of temperature x
appears to be approximately a function of depth, H, alone if we define t
it as
\ = °-5 (*o Km>
- 0.3 (V/3 <
TT
2H
(2.7)
Fig. 3 is a graph of x^ calculated from Grant's (1968) values of c and H, as
well as those of Nasmyth (1970).
O GRANT etal (1968)
X NASMYTH (1970)
20
40 60 80
DEPTH, H METERS
00
Fig. 3 Transition wave numbers, calculated from Eq. (2.7) using data of
Grant (1968a) and Nasmyth (1970). The empirical fit to the data,
given by Eq. (2.8) is shown.
.The equation of the best simple approximation to this graph, is of the form
-H/h
x = B e
(2.8)
where B = 9 .0 m-1
h = 40.0m
H = depth, m.
x = lower wave number limit of isotropic
temperature spectrum, m
12
For example, for the experiment at 27M depth, shown in Fig. 2, Eq. (2.8)
allows us to diagnose that the medium was no longer isotropic for x (x) >bx ' for x* ^x^x , and a
rapid exponential drop-off for x>x . We will generalize and use this spectrum
in the next section. The data available show that x «x. «x .
o t m
2 .2 The Spatial Correlation and Structure Function
The alternative to the spectral description of fluctuations of temperature
or index of refraction is either the correlation function or the structure function,
13
which is simply derived from the spatial correlation. These functions turn out
to be the most direct way to calculate the sound fluctuations , as we will see
in Section 3 .
One of the earliest determinations (Liebermann 1951) of the spatial
correlation of temperature fluctuations was conducted by mounting a therm-
istor on a submarine operating at a depth of approximately 50m. The
standard deviation of the temperature fluctuations was found to be approxi-
mately 0.4C at this depth (time of day, weather conditions unreported).
The spatial correlation could be fitted approximately by the exponential law
(shown by solid line in Fig. 5)
cT <§)
2
= e
-S/a
(2.9)
where a = 60. cm
which thereby defined a correlation length, sometimes called a "patch size
of 60 cm.
i.o k^
\:i
Kolmogoroy
N^_ 0 ■*•••
100 120
E cm
Fig. 5 Temperature correlation function from Liebermann (1951) with data
(circles), empirical fits to exponential (solid line), and Gaussian
(dashed line, correlation functions. The correlation based on Fig. 4
with x = 0.24 cm is shown by dotted line, identified as "Kolmogorov1
Others have used the Gaussian correlation function
2
CT (4) =e
-(£/*)'
(2.10)
to describe micros tructure fluctuation. A Gaussian correlation function
(dashed line) has been drawn in Fig. 5 for comparison with the exponential
functions and the data (circles).
14
Both functions define the same correlation length, since they both reach
e at £= a. The correlation function of the index of refraction will equal
that of the temperature, if other parameters affecting the speed of sound
have negligible effect, and if the fluctuations are small. Then
cT(*)=cu («>
It is not always possible to determine the correlation function C (£) or
C (4) by measuring it directly. In any event it would be a boon if we could
use the known universal characteristics of ocean micros true ture to deduce
the spatial correlation function in the real world outside of the laboratory.
Although this turns out to be not completely feasible, there are certain
generalizations about the spatial correlation function that can be deduced; one
of these is its form at small values of £.
The spatial correlation function of the excess index of refraction can be
obtained from the wave number spectrum by the Fourier transform theorem. In
three dimensions this relation is
s(?)-"-f- ^ y*)ei*'* dH (2-n)
where f* is the vector spatial lag.
If we now assume that the changes of the index of refraction are independent of
the direction being studied we can simplify the integration. In spherical
coordinates
H»r. = urease- and dn = 2tth sin©- dn d &
Then C (r) = -*Z— fQ K2 dH j" s (h) sin@r ^nr*** rf
a ^
and the £ integration yields
We now define the one dimensional spectral function, $ (x) for the isotropic
medium
"• 2
S^ (h) dn = 4tt h S^ (h) dH = § (h) dn and change (2.12)
notation, r -» 4 # for consistency with previous work, so that
15
% (§) = —\- £ li£ri V (K) dH- <2-13> ■
£ may be in any assigned direction for the isotropic medium.
In order to integrate (2.13) we start with the spectral conclusions of the
previous section (Fig. 4). We are prepared to make four simplifying assump-
tions in order to proceed:
1) $ (H) = $ + s(H - K ) for K < K , : *
M- m my ra (2.14)
$m;-$o Source Subrange
where s is the slope constant= xm
$m = peak isotropic PSD; $0 = $ (°)
2) * 00 = constant = 4 for * < * < ** Transition Subrange
K -5/3 ra x
3) §^ (h) = $m ( t/H) for Ht < k < Kq Inertial Subrange
4) $ 00 = o for H > H0 Dissipation Subrange
The approximate spectrum that follows these assumptions is shown by the
dashed line of Fig. 4.
The effect of the simplifying assumptions is to replace the highly variable
anisotropic range and the fixed, but complicated, isotropic range by a fixed,
easily-integrable , spectrum. This will permit us to assess the relative impor-
tance of each of the sub-ranges and to determine what universal generalization
can be made, (if any) about C (£) at sea.
Using the assumed spectrum of Eq. (2.14) it is possible to immediately
determine the excess local index of refraction, a in terms of the peak spectral
density, $ , and the constants x and x . Recalling that the mean squared value
of any fluctuating quantity may be obtained by integrating the spectral density over
all values of the appropriate parameter (frequency or wave number) we find
(V00
(u ) = J $ (K) dK
^r O JJ,
m m t
Since x « x and & < & , the result of the integration simplifies to
m t o m
(j, = 5/2 k $
t m
16
It is interesting to determine the parts of the spectrum most responsible
for this variance of the fluctuations . The terms in the integration were
1/2 x <£ source subrange
m m
x, $ transition subrange
t m
3/2 xx $ inertial subrange
t m
Since x « x we conclude that the contributions to cr come almost equally
from the transition and the inertial subranges. Furthermore, our simplifying
assumption of the particular form of the source subrange can have little effect
on (j because that contribution is proportional to x , and x «x .
\i m m t
A second general conclusion can be reached if we restrict our interest to
very small values of the spatial lag, £ . Then we can substitute the expansion
for the sin x£ and integrate Eq. (2.13) with ease
— Jo $ (h) dx a__ (J H [$ + m (h - x )] dx
+ Jk A
m m
§ dx
nv m
+ jHoX2J (^)5/3dK
K m x
2
The first integration yields (\i >, as we saw above, so that the first term
is unity. The integration over the three terms in the curly bracket is also simple
so that
.2 . 5/3 4/3
C (§) = 1 - % $m Kt ^—
r2 (2.16)
8 o-
and with the aid of (2.15)
C (§) = 1 - 0.05 k 2/3 x 4/3 §2 for x 6«1. (2.17)
fJL tO O
The significance of this result appears if we consider any specific depth ^ say
50m. Then (2.3) gives x = tt/10 cm" , (2.8) predicts x = 2.4 x 10~ cm"
and therefore, at 50m depth, we can expect
C (§) = 1 - O.J468 §2
17
For small displacements this is an expansion of the Gaussian form
C^ ' V- o SlvEH § *« + S (H - H )]dH + J ^V^»'dH + J °sAlLHi* ^5/3 ,
M- U H = o H§ u in v m/J h h£ m k hZ m x
m t
The coordinate for integration is now changed to 0 = x £
to « ■ i/:l -&**\ ♦ -f - *» * h:! 5<^>v ♦ ?2i:^*/H
m 3 '
s in R R £i
The approximations are now made for J p d/S = |9- -~r-, + 5°5 , in the source
integral and in the transition integral. Because of the rapid convergence due
to the jg 8/3 in the denominator, the third integral is comfortably approximated
by using the sine expansion from zero to tt/2 and again from ir/2 to n. Using
(2.15) the result of these integrations, valid for £ £— with about 5% error, is
\
C (i) = 1.0 - 0.48(xt£)2/3 +0.28 (x^)2 +6.7 x 10~4(xt£)4 (2>18)
The curve is shown in Fig. 6
I.O-
*-*
\j 0.8-
v:
3. a6-
^ 0.4-
0.2-
.
0
i
i
i
i
i
i
w
0 05
1.0
1.5
2.0
2.5
3.0
M
Fig. 6 Spatial correlation function C (x £) derived from spectrum of Fig 4
It turns out that the source subrange makes essentially no contribution to
the correlation function for 4 = — J dx = — J dx
c o c o
o o
because the average
index, (n) , is unity.
20
It is the deviation from the mean that we must calculate:
At = t - =r — J n (x,T) dx — J dx
CO CO
o o
i rR
= — J0 M.(x,T)dx
where p, (x,T) = n (x,T) -1
From this we get the square of the deviation from the mean (At) =^ — „)
( Jo |jdx f^dx) which can be written (At) =— J^ J^ |j (x^T)^ °(X2'T)
o
dxldx2
2 1 R R
The mean square deviation from the mean is (At > = - dxn f " (u (xn ,T)
/m\\_i cz«Jol«Jol
[i (x2,T)> dx2 o
The integrand can be written in terms of the spatial correlation of the excess
index of refraction C (£):
= (n*> C^ (x2-Xl) = a* C (VXl) .
a2 „ /xR
2 R I
In these terms (At ) = — ^ J C (x^-xjdx,
0 o ^ / 1 2
v p *. 4. (3.3)
o
We seek the mean squared phase fluctuation which is calculated by
substituting cp = ceAt where (pis the instantaneous deviation of the phase from
the value due to the mean speed, c . The units of =a2k\fc (4) d£ D«l (3.4)
2 2 2 2
Since (cp } = (cp> + Var cpand (cp> = o, by definition, (cp )= Var (p= a . (3.5)
We have only to know the spatial correlation of the index of refraction, from
the source to the receiver, in order to carry out the integration, and obtain
the variance of cp .
2 2
It should be pointed out that the quantity a = ([± > must be known at the
time, and location of experiment. The significance of this comment is that, in
the ocean, the evaluation of a can be strongly dependent on position and the
range at a given time, or a function of the time and duration of measurement.
21
For example during passage of an internal wave, at positions in the thermo-
cline there will be large variations of the temperature and the speed of sound,
over times of the order of several minutes.
It is also convenient, sometimes, to assume that the spatial correlation
function drops to zero after a few correlation lengths in either direction from
any reference point. Then for the sound path of length R »a we can extend
the integration to infinity in both directions , or double the integration from
zero to infinity, without affecting the total integration,
00
a2= ?1 as a requirement
for far field propagation, the fluctuations of the medium, a must be so small that
2 2
k* a. aR « x
In practice this condition is not difficult to fulfill; field measurements
generally show q <10
22
The Gaussian correlation function
C (£) <=e "^/a) (3.8)
is attractive because of the simplicity in evaluating the integrals of Eqs . (3.6)
and (3.7). For the Gaussian correlation function the phase fluctuations are:
a2 = = ^CT2 k2 Ra D«l (3.9)
2 . 2V y? 2 2 (3.10)
and a = < =-r— a k R a D» 1
= a2 K2 Rfc (£)d£ D»l (3.14)
A (_i Jo (j ^ ^
where cr is the fractional standard deviation of the pressure amplitude
o = <(P-PQ >2)% (3.14a)
A Po
23
Comparison of expressions for amplitude fluctuations with those for phase
fluctuations is revealing. For example, for the high frequency situation D «1
the mean squared amplitude fluctuations depend on an integration of the fourth
derivative of the correlation function in the transverse directions, y and z,
rather than a straightforward integration over the correlation function in the x
direction as in Eq . (3.8) for the phase shift. This means that the amplitude
fluctuation is caused by the curvature (really double curvature) of the correla-
tion of the refractive index, transverse to the sound path. The dimpled medium
in the case of amplitude fluctuations therefore behaves as a series of lenses
coverging or diverging acoustic energy along the sound path (notice the evalua-
tion at zero displacement in the y and z direction, that is, along the beam).
At the other extreme of propagation parameters, however, when the range
is larger, or the frequency is lower, such that R/a » ka (D »1), the sound
field fluctuations are caused by field distortions due to a large number of
inhomogeneities along the path in the medium; these distortions interfere,
constructively and destructively along the entire path and this interference
effect is common to both phase and amplitude. For this condition the mean
squared fluctuations of phase and amplitude are identical. Both are linearly
dependent on: a) the distance of propagation, b) the frequency squared, and
c) the integrated spatial correlation function.
It must be pointed out that all of the derivations to this point have been
based on the assumption of single scattering by weak fluctuations. The
consequent solutions are valid only in so far that very long paths and very
large fluctuations are not allowed. In practical propagation over very long
paths the magnitude of the phase and amplitude fluctuations cannot increase
with R without limit, or they would reach the absurd conditon of fluctuations
that are greater than the quantity itself. In fact, "saturation" must take
place and the increase of the fluctuation with range must reach a limit that
will be indicated, shortly.
If the Gaussian correlation function of the index is assumed (Krasilnikov
195 6) the variance, ah, is:
°l - C-^fV) k' R a <1-^-tan"1 D)1 (3.15)
al (X2> = (-^T") °l (R/a>3 D = (~T~) °l ^ (Ra) D >;> l (3'17)
24
3. 3 Small Amplitude Fluctuations: Laboratory Experiments
2
The dependence of small values of or on the parameters of the medium and
the frequency has been put on a firm experimental basis in a beautiful set of
laboratory studies (Stone and Mintzer 1962) in which the micros tructure of the
index of refraction was obtained by heating the tank of water through which
the sound beam was propagated. First the rms value and the spatial correlation
function of the temperature microstructure were obtained by direct measurement.
From these data the value a / and the correlation function, Cyi (£ ) , of the
excess index of refraction were calculated. The spatial correlation function was
fitted to the Gaussian form, and the correlation length, a, was calculated from
Eq. (3.8). The measured temperature micros tructure , fitted to the Gaussian
correlation function showed a correlation length, a = 3.5 cm.
Additional comparison showed that the rms excess index of refraction,
deduced acoustically, was approximately 25% less than the value 1.6 x 10 ,
calculated directly from temperature measurements. One flaw was revealed
in the experiment: the acoustically determined value of the assumed Gaussian
correlation length was only about one-third of the value (3.5 cm) directly
measured by thermistors. This discrepancy was possibly due to the error of
assuming that the correlation function was Gaussian. As shown in section 2,
the Gaussian form is appropriate only for very small displacements. A
combination of Gaussian and Kolmogorov correlation functions would probably
have been a better description of the medium for insertion into Eqs. (3.13) or
(3.14).
3. 4 Large Amplitude Fluctuations - Large Ranges
So far we have considered small fluctuations for two cases:
1) Change of sound pressure amplitude, P, in the presence of zero
(or negligible) phase shift. In section 3.2 we gave expressions for
2 P -P 2 2
cr. = // o_. v . The type of pressure field at x = R that results in = 0;
S o
b (x,t) = time varying magnitude of the scattered pressure which is 90 out-of-
phase with the original wave at x. We assume b is Gaussian distributed and
= 0. b *A.
s s
Equation (3.18) resolves the fluctuation scattered pressure, P of Fig . 7 into
components in-phase with the depleted original wave A (x) ana out-of- phase
with it. A typical situation in Fig. 7 shows that
«2 2 /I
P = a + b
s s s
To get the typical amplitude, Pf of the total wave at x as a function of A,
a , b , Eq. (3.18) is rearranged to
P = (A + a ) cos (cot - kx) + b sin (wt - kx) (3.19)
= P cos (o>t - kx - $)
where P = pressure amplitude at x (3.20)
r/» x2 u 2-,l/2
= [(A + a ) +b ]
s s
<$ = phase angle with respect to original
-1 bs
wave at x = tan ( ) (3.21)
A+a
s
It is useful to expand _P and $ for small fluctuations a — b «A.
Using the binomial theorem for _P
-A(l +-5"- +%A1N )-A+as (3-22)
and the small angle approximation for $
I = tan"1 (-^- ) -
A+a ; A (3.23)
s
27
2
It is therefore clear that for small fluctuations the = < P2)
UJ > = <(A4.s)2 + b/)
■= + (2ao A) + (a 2> +
_ 9 9 9 ^ ^
Ao " * + (as > + (3.28)
To determine the attenuation rate, a , we use (3.2 7) and (3.28) and form
S3
An2 -A2 _ (as2) + (bs2> _ , -20i R
A 2 " A 2 " L "e S (3.29)
A A
o o
The evaluation can be made in the small fluctuation range where the
ingredients of (3.29) are well known. There, using the expansion of the
exponential for small argument and (3.24) and (3.2 5)
aA + a9 = 2asR (3.30)
The condition for long range is of particular interest.
Then we have, from (3.7) and (3.14)
°l = al >D>>1 <3-31>
so that the attenuation constant of the original beam due to energy going into
fluctuations , is
* =a2/R ,D» 1
(3.32)
In particular, when the scatter is due to a medium with a Gaussian correlation
function, C (£), we have, using (3.17),
a =-^-a2K2a (3.33)
s 2 |i
29
We can now proceed to consider large ranges and large fluctuations.
The quantity observed in any experiment is the pressure amplitude P. Eq. (3.2 0)
shows P to be a function of all three quantities: The steady coherent component,
A, and the fluctuating random components, a and b . The conventional way to
express the amplitude variations is in the ratio
(CAV)2= -2
2 (3-34)
2
where CAV = coefficient of amplitude fluctuations. It might be noted that for
small fluctuations we have simply CAV = a., Eq. (3.14a). Here, for large
fluctuations we find, from the conservation of energy assumption that
= = Ao2 (3.35)
2
But the calculation of involves a more involved evaluation of the mean
value of P given by Eq. (3.2 0) when both a and b vary randomly, each with an
assumed Gaussian probability density function.
The mean value of the sum has been found by Brownlee (1973) and is
=^AQ [1 -e-2asR}HexP[-2/(e2^R-i)]}^ £
^ i •>, , (3.5)b2 (3.5 7)b3 (3.36)
where £ = 1 + Jb + ^, .j. + tjj v* +
b « fcCe2** -D"1
2 2
When this is now squared and combined with P in Eq. (3.34) the A cancels and
the result is expressible solely in terms of the attenuation constant, a , and the
length of path, R.
(CAV)2 =-1 +-^exp ( ,l„ m . ) {[1 -exp (-2« RJIS2]}"1 (3.37)
77 exp(2asR)-i s
This messy expression has a simple maximum value for large fluctuation and
large ranges; we get
(CAV)2 = -1 + -4" = 0.273 for 2a R » 1
v tt s (3.38)
or (CAV) = 0.522
max
The complete behavior of the coefficient of amplitude variation is shown in
Fig. 8.
30
VM
i
-1-1
4-J
d
o
fO
3
4->
i—4
o
3
CD
1 — 1
T3
>4-l
00
3
4->
-•-1
a)
d
a
H
£
(0
C
o
O
IS
c
-C
o
0
•r-4
CO
4->
OJ
TJ
S-4
-1-1
(0
>
N
T3
ffi
(0
-a
^
d
LO
-a
CM
c
4->
m
rd
4->
CO
r^
0)
CD
to
jC
cr>
6S
4->
4-4
r— 1
O
O
c
,*
0
£
d
-1— 1
4->
o
d
4-1
of Shvac
ure only
IO
co
C
data
truct
d
o
-r-4
4->
CO
cD o
x: u,
H O
1-4
-i-H
rd
. S
>
^o
" CD
co 3
-I-H
c -o
i— 1
o
a,
m-j cD
6
« -Q
rd
3 0
^■M
<-M
o *-
d
O
4->
— i CD
C
0)
•^H
o
fO co
■1-1
C CO
4-4
o
:>
a)
0
O
GO
•
Cn
PL,
o b
31
which is the graph of the square root of (3.37). We observe that the fluctuations
(CAV) = a for small values of a R, that is, for small fluctuations. The linear
dependence of (CAV) on range continues until approximately a = 0.35 after which
growth slows. (CAV) reaches 90% of its ultimate value at approximately
VK (J2
M
can be determined for any spatial lag, the value of C (£ ) at £ >R is quite
irrelevent to a calculation of the sound fluctuations for a path of extend, R.
Therefore practical acoustical application of integrals of C (£)d£ will be
satisfied by an upper integration limit, R, rather than °° .
The microstructure study of section 2.2 leads us to have confidence in the
form of C (£) only for spatial lags over which the Kolmogorov spectrum is
dominant. We have set these limits as 0 <£
o
o
o
x: jo
U
CD ^_,
CO
o
(D
x:
t*»
w
o g
6 c
a p
CD
CD
w
(0
en
CD fd
fd
CO
0
S-.
p
-M
u
p
ii
CO
O
S-.
O
XS
CD
+->
fO
o
c
p
+J
CD
Xi
+j
CO
•iH
CD
p
T3
CD
o
o cp
r-> CD
O
T3
CO
6 H
CD
> o
. (D
0> g
Q_ lO
^ en
C-> _e
2 *
c
CD
£
CD
3
en
fO
CD
c 6
.2 -
-t-> fO
CD CD
o
(D
a
en
>
O
cn
o .2
£ -3
r— I (0
O ^
-S o
o
o
CD
^-. a
co X
j3 (D
CO Z!
CD
C
(0
-rH
en
en
P
fO
,Q O
•■H
33
it levels off while the actual correlation function shows a continuing decrease
and then an oscillation with small amplitude around the zero correlation axis.
In order to obtain the integral of C (£) for a propagation experiment in such
a sea we have two choices: One solution would be to determine C (£) and
C (£) out to the range of the acoustic experiment; this would not only be a
formidable task but unrealistic for any long range experiment. The second, more
expedient, choice is to accept a judicious combination of theory and empiricism.
We propose to use the theoretically based microstructure of Fig. 6 to determine
the correlation length, a, and then to use the Gaussian correlation function to
approximate the true correlation curve. Judging by Fig. 9, this will result in
slightly too large a contribution for JC (£)d£ between 4 = 0 and £ = 0.5M. We
will assume that subsequent positive and negative parts of C (£) will cancel
each other and contribute nothing to J*C (£)d£ from £=0.5Mto£=R. In the
example being discussed, the value of rjhe correlation length, a = 25 cm.
Having found a suitable way to determine the equivalent correlation length,
all of our algebra for the sound fluctuations due to a Gaussian correlation func-
tion is assumed to be applicable, even though the true correlation function is
by no means Gaussian.
The general method for the determination of the correlation length is found
by referring to Fig. 6. The correlation function is down to e at
x £ = x t a = 1.8 (3.39)
therefore for our approximation
1.8
a =
xt
Furthermore, using the empirical approximation to the transition wave number
that we developed from Fig. 3, Eq. (2.8), we have the generalization for the
upper ocean.
a = 0.2eH/4°
where H = depth, m (3.40)
a = temperature (or index) correlation distance in m.
It is evident from the data scatter of Fig. 3, from which the constant
B and H are determined, that (3.40) can be only a guideline for the prediction
of the correlation distance.
34
Although a often decreases with increasing depth, a general rule describing
(j as a function of depth is not possible at this time. In fact, when such a rela-
tion is found there will certainly be other important parameters in the equation,
such as time of day, to which a is particularly sensitive at shallow depths.
When a is measured at the place and time of an acoustics experiment,
Eq. (3.40) provides a value of correlation length, a, that completes the needs
for evaluating q or a . For example, Shvachko (1967) has made measurements
of a and CAV af depMs 20, 35, 40M using sound of frequency 25 kHz. If we
take Schvachko's value of a / R/ k/ and use his depth in Eq. (3.40) to obtain
a, we can calculate q from (3.17) and predict his CAV from (3.37) or the graph
Fig. 8. This has been done in the following way:
ILLUSTRATIVE EXAMPLE
Shvachko (1967) has performed an acoustic fluctuation experiment at
frequency 25 kHz, range 20 m, in the sea of Norway. The medium had a
standard deviation of the excess index of refraction n = 17.8 x 10 . We are
to calculate the predicted cr .
At b. /depth 20 m reading from Fig. 3 x = 0.055 cm"" or using (2 . 8) x =
9.0e~ = 5.45 m~
1 8
therefore from (3.39) a = — '— = 0.33 m
t
20/40
or directly from (3.40) a = 0.2 e ' = 0.33 m
CO 2 (25.x 103) ._. e -1
k= — = -rh lnA 104.6 m
c 1.5 x 10s3
calculate D= 7—2 n i. cl, 0?2\ = 84« >>l
ka^1 = (104.6)(.33^)
therefore, using the Gaussian assumption (3.17) for D »1
A 2 u
= (0.89) (17.8 x 10"5)2(104.6)2(240.)(0.33)
= 0.0244
oA = 0.16
This is to be compared with Shvachko's CAV = 0.17
35
Since the