HYDROACOUSTICS AND

UNDERWATER ACOUSTICS

SESSIONS

New Virtual Sonar and Wireless Sensor System Concepts

J. A. Bucaro, B. H. Houston, and A. J. Romano
Naval Research Laboratory, Washington, D.C. 20375
Recently, exciting new sensor array concepts have been proposed which, if realized, could revolutionize how we approach
surface mounted acoustic sensor systems for underwater vehicles. Two such schemes are discussed here so-called "virtual
sonar" which is formulated around Helmholtz integral processing and "wireless" systems which transfer sensor information
through radiated rf signals. The "virtual sonar" concept provides an interesting framework through which to combat the
deleterious effects of the structure on surface mounted sensor systems including structure-borne vibration and variations in
structure-backing impedance. The "wireless" concept would eliminate the necessity of a complex wiring or fiber-optic external
network while minimizing vehicle penetrations.

INTRODUCTION
The growing possibility of being able to implement
acoustic systems with high sensor counts ( 10 4) has
motivated consideration of how we might exploit
such a capability when it does indeed become a
reality. In particular, fiber optic sensor arrays
utilizing in-fiber Bragg gratings and the revolution
underway in MEMS/NEMS silicon-based sensor
technologies suggest that such high sensor count
systems might be just around the corner. Our
considerations of how one might exploit this future
technology for underwater vehicle sonars has
centered on long-standing technical and engineering
issues which have hampered these applications.
These are, first and foremost, the deleterious effects
of structure-borne noise and hull impedance spatial
and temporal variability. But they include as well
limited apertures and the necessity for multiple hull
penetrations for sensor signal feed through. Two new
sensor array concepts one called virtual sonar[1]
and the other involving wireless arrays based on
cellular communication[2] offer a new perspective
on how to approach surface mounted acoustic sensor
systems in the attempt to mitigate these problems.
The "virtual sonar" concept provides an interesting
framework through which to combat unwanted
effects of the structure on surface mounted sensor
systems. The "wireless" concept would eliminate the
necessity of a complex wiring external network while
minimizing vehicle penetrations. In the following, we
discuss these concepts and how they might be
synergistically applied in the case of an underwater
AUV.

ARRAY CONCEPTS
Virtual Sonar
The virtual sonar concept[1] is based on simple
considerations of evaluating the Helmholtz integral

FIGURE 1. Helmholtz integral reconstruction for an

incident plane wave, point force, and the two combined.

over the surface of an underwater structure for which

there is high spatial density acoustic pressure and
normal velocity sensor data. The important result
here is that for the case of a structure excited by both
an incident acoustic signal and interior noise sources
applied to the hull, when the acoustic field is
evaluated inside the surface of the structure, only the
virtual incident field remains. This is illustrated in
Figure 1 taken from Reference[1] for an evacuated,
thin cylindrical shell at kaa = 5, where ka is the
acoustic wavenumber and a is the shell radius. To
illustrate the structural noise reducing properties of
virtual sonar processing, Figure 2 displays the
spatial Fourier transform of the pressure over the
length of the cylindrical shell section for this case.
The lowest curve is the transform of the virtual sonar
evaluated along the central axis, as well as for the
true 1 Pa incident field (these two curves overlap).
The upper curves are the transforms of the total
surface response along a line on the lateral surface of
the cylinder when the interior force (normalized by
the square of the shell thickness) is 104 Nt/m2 and 10 3
Nt/m2, respectively. One can see the superior
performance of the virtual interior sonar regarding

SESSIONS

structure-borne noise across the entire wavenumber

spectrum including the acoustic domain (-ka to + ka).
10000

104 Nt/m2
1000

Exterior Line Array

103 Nt/m2
100

Virtual Sonar

10

FIGURE 3. Wireless array on an AUV

-ka
0.1
-20

-15

-10

-5

ka

10

15

20

Wavenumber

FIGURE 2. Spatial Fourier transforms of the pressure over

an exterior line and for the interior virtual line. The
excitation is an incident plane wave and an interior point
force. The transform for the incident field overlaps that for
the virtual sonar.

Results of the type shown above require sensor

spatial sampling on the order of two per structural
wavelength. Azimuthally, the number of sensors, N,
would be 2nmax, i.e twice the highest flexural
circumferential harmonic. Axially, N would be
2L/f, i.e. twice the number of flexural wavelengths
(f) along the shell of length L. Now nmax (k aa)2/5
and f (k a a) 1/2. Calculations for the flexural wave
dispersion and cut-off frequencies versus nmaxfor a
plastic-like, 1.8m AUV shell structure predict the
following. At kaa = 1, about 103 measurement points
would be required, while for kaa =10, about 8600.
This requires indeed a large number of sensors;
however, it would enable a noiseless array. In
addition, with both pressure and velocity
(acceleration) sensors employed, hull impedance
spatial and temporal variations would be of no
consequence.

Wireless Array
One approach for accessing such a large sensor
count system is a wireless array. As depicted in
Figure 3, wireless sensor/radios would be distributed
over the surface of the structure. As many as one
thousand sensor/radio pairs could be tied in to a
specific cell cite, and there could be tens of cells over
the body. The sensor information from this relatively
small number of cells could be fed onto a single line
which then penetrates the hull. The two major
technology issues here involve propagation of the
radiating gigahertz rf signals from sensors to base
cell and powering of the individual sensor/radio
devices.

In the case of a small AUV, an individual cell

might access sensors located within approximately a
half meter radius. At the high rf frequencies
involved, the high absorption in water necessitates
the introduction of a suitable waveguide material to
allow sufficient propagation even over these modest
base-cell distances. The field strength of the lowest
TEM order mode is approximately proportional to the
following factors
G c/ 2

fre

r tan

(1)

where c is the speed of light,

is the dielectric
constant, f the frequency, r the propagation distance,
the permeability, and
the loss factor. Eq. (1)
indicates that what is desired is a material with a low
index of refraction and a low loss factor. In water,
the rf signal decays by 10-5 in 1 cm. This compares
to 2m, 12m, 20m, and 1km for polyurethane, nylon,
Teflon, and Styrofoam, respectively. Thus, a thin
layer of a material of this type would provide a
sufficiently low loss waveguide in which to
propagate the sensor-to-base cell signals.
Estimates indicate that on the order of 1.5mW of
electrical power would have to be supplied to each
sensor/radio pair assuming 10nJ/bit. This could be
distributed in a number of ways including power
broadcasting and energy harvesting of thermal
gradients or fluid flow.

ACKNOWLEDGEMENT
This work is supported in part by ONR.
REFERENCES
1. A.J. Romano, J.A. Bucaro, B.H. Houston, and
E. G. Williams, J.Acoust. Soc. Am. 108, 2823-2828
(2000)
2. A. Mehrotra, Cellular Radio: Analog and Digital
Systems, Artech House, Boston, MA, 1994

SESSIONS

Development of Thin, Low Frequency Electroacoustic

Projectors for Underwater Applications
T. R. Howartha and J. F. Tresslerb
a

NAVSEA Division Newport, Newport, RI USA

Naval Research Laboratory, Washington, DC USA

Two acoustic transducer panels have been designed, fabricated and electroacoustically evaluated. These panels featured
cymbal drivers sandwiched between a radiating cover plate and a tungsten backing plate. The acoustic output shows
resonance frequencies of both transducer panels below 1 kHz.

INTRODUCTION
The two most common acoustic projector technologies
used on unmanned underwater vehicle (UUV)
platforms are tonpilz transducers and piezocomposites.
Both of these technologies, however, are typically
designed for use at frequencies above 10 kHz. A U.S.
Navy designed 1-3 piezocomposite 2.54 cm in height
with a projector radiating face of 15.24 cm by 7.62 cm
was demonstrated to exhibit broadband characteristics
between 10 kHz and 100 kHz [1]. At its resonance
frequency of 100 kHz, the TVR was measured to be
174 dB//mPa/V @ 1 m. At 1 kHz, the TVR was less
than 110 dB//mPa/V @ 1 m.
To generate high acoustic output at frequencies below
10 kHz, free-flooded piezoelectric ceramic rings,
electromagnetic drivers, and flextensional transducers
have traditionally been used. However, due to their
large size and weight, these technologies are not easily
adaptable or convenient for use in UUV platforms.
For the implementation of low frequency acoustic
sources on UUV platforms, advanced hardware is
required. The U.S. Navy has designed, built and
evaluated novel prototype underwater electroacoustic
projectors that have a fundamental resonance
frequency below 1 kHz. The active component of
these projectors is slightly less than 0.5 cm in height
and has a radiating face of 15.24 cm by 7.62 cm,
making them ideal candidates for use in mobile
platforms and littoral environments.

Materials Research Laboratory at Penn State in the

mid 1990s and have been investigated for use in a
number of applications [3]. A cymbal consists of an
electroactive ceramic disk sandwiched between and
mechanically bonded to two thin metal caps. Each cap
is shaped in a die press so that it will contain a shallow
air cavity underneath its inner surface after it is bonded
to the face of the ceramic disk. The caps serve as
mechanical transformers for converting the small
radial displacement and vibration velocity of the
electroactive disk into a much larger axial direction
displacement and vibration velocity normal to the apex
of the caps. Hence, the cymbal driver primarily
utilizes the 31 contribution of the active ceramic to
achieve flexure in the caps.
Cymbals with two different diameters (12.7-mm and
15.9-mm, designated as Type 1 and Type 2,
respectively) were used in this study. Titanium was
selected as the cap material because of its low density
(4500 kg/m3), moderate elastic modulus (120 GPa),
and oxidation resistance. The electroactive material
was Navy Type VI (PZT-5H) piezoelectric ceramic.
This material was selected because its very large
piezoelectric d31 coefficient served to generate the
largest flexure (i.e., displacement) in the caps.
Prior to bonding the caps to the ceramic, studs 1.4 mm
in diameter and 4.6 mm long with UNF 0-80 threads
were welded to the apex of the caps to form both the
mechanical and electrical means of handling.

ACOUSTIC TRANSDUCTION PANEL

THE CYMBAL DRIVER
The low frequency underwater acoustic transducers
utilize miniature class V flextensional drivers [2],
commonly known as cymbals, as the active elements
in the projector. Cymbals were developed at the

The low frequency acoustic projector panels consisted

of an array of cymbal drivers sandwiched between a
12.7 mm thick tungsten backing plate and a 2.2 mm
thick copper electroplated carbon graphite epoxy board
(solid all uni-carbon from Aerospace Composite

SESSIONS

Products, San Leandro, CA). Figure 1 shows the

cymbal elements mounted into the tungsten backing
plate before the cover plate was attached.

driven to a maximum of 750 Vrms because further

evaluation was desired and we didnt want to take the
risk of failure. Comparing the transmitting response of
Xducer I both before and after being subject to a drive
level of 750 Vrms indicated no adverse effects due to
the high drive.

150
12.7-mm diameter driver (solid line)
15.9-mm diameter driver (dashed line)

FIGURE 1: Cymbal elements on backing plate.

The total radiating area of the projector was 152.4 mm
by 76.2 mm. Two transducers, designated as Xducer I
and Xducer II, were built and tested. Xducer I
consisted of 50 Type 1 cymbal drivers in a 10 by 5
arrangement. Xducer II contained 32 Type 2 elements
in an 8 by 4 configuration. The studded cymbals were
initially torqued into a predrilled and threaded tungsten
backing plate. Next, the drilled out graphite cover
plate was lowered into place so that it rested over the
top of the cymbal array. The cover plate was similarly
torqued onto the studded cymbals with hex nuts.
Standard underwater calibration measurements on the
low frequency projectors were performed at the
NAVSEA Division Crane Glendora Lake Facility in
Sullivan, Indiana. The measurements were performed
at a depth of 11.3 meters and a water temperature of 10
degrees C. The transmitting voltage response (TVR)
for the transducers as a function of frequency was
measured from 0.5 kHz to 100 kHz and is shown in
figure 2. The low frequency resonances, 0.6 kHz in
the case of Xducer II and 0.9 kHz in the case of
Xducer I are a result of the piston-like motion of the
cover plate being driven by the individual cymbal
drivers. Both projectors exhibit markedly higher TVR
than that of a 1-3 piezocomposite with the same
radiating area.
Unlike the 1-3 piezocomposite,
however, the cymbal-based transducers are
multiresonant as the cap and flexing graphite plates
flex throughout the frequency bands. The beam
patterns at 1 kHz for the two projectors both show the
omnidirectional response as is to be expected at these
frequencies. The sound pressure level (SPL) was
measured for Xducer I as a function of frequency and
drive level. Nonlinear behavior becomes apparent at a
drive level above 500 Vrms. The transducer was only

TVR (dB//uPa/V @ 1 m)

140

130

120

110

100
0.1

10

100

Frequency (kHz)

FIGURE 2: TVR comparison of transducers.

ACKNOWLEDGMENTS
The authors express their appreciation to Walter
Carney, Kirk Robinson and Mel Jackaway of
NAVSEA Crane Division for fabrication and
measurements of the panels. The authors acknowledge
the support of the Office of Naval Research.

REFERENCES
1. T.R. Howarth and R. Y. Ting, Development Of A
Broadband Underwater Sound Projector, CD
Proceedings of OCEANS 97 MTS/IEEE Conference,
IEEE
Publications
(ISBN
0-7803-4111-2/97),
Piscataway, NJ, 1-7, 1997.
2. R. E. Newnham and A. Dogan, Metal-electroactive
ceramic composite transducer, U. S. Patent 5,729,077,
issued March 17, 1998..
3. J. Zang, W. J. Hughes, P. Bouchilloux, R. J. Meyer Jr., K.
Uchino and R. E. Newnham, "A Class V flextensional
transducer: The Cymbal," Ultrasonics, 37, 387-393,
1999.

SESSIONS

Experimental Results of Passive Phase Conjugation Applied to

Underwater Acoustic Communication
W. L. J. Fox, D. Rouseff, D. R. Jackson, and C. D. Jones
University of Washington, Applied Physics Laboratory, 1013 NE 40th St., Seattle, WA 98105, USA
In May 2000, an experiment was performed to demonstrate the effectiveness of passive phase conjugation (PPC) for underwater
acoustic communications. PPC is a coherent array and signal processing scheme that has strong theoretical ties to active phase
conjugation and time-reversal mirrors in the ocean. It differs from active techniques in that an array of transducers need only receive,
while the transmitter can be a single transducer. Previous results have been presented showing performance of the algorithm for cases
with a vertical receive array spanning the water column. This paper will show further results with a truncated receive array.

THEORETICAL FOUNDATIONS OF
PASSIVE PHASE CONJUGATION
Phase conjugate acoustics have previously been
demonstrated in the ocean [1]. The basic phenomenon
can be described by the following. A source transmits
a signal, a distorted version of which is received by the
transducerss of a distant vertical array. The signal is typically distorted by time spread due to multiple interactions
with the sea surface and bottom as it propagates down the
acoustic channel. If these distorted receive signals are
time reversed and transmitted from their respective transducers, it can be shown theoretically [2] that there will be
a spatio-temporal focusing of this second transmission at
the location of the original transmission.
It is the time-spread distortion mentioned above and
its time variable nature that pose the main hurdles to high
data rate underwater acoustic communications. Computationally complex adaptive equalizers are the standard
technique for addressing these problems. A technique
called passive phase conjugation (PPC) has been developed and presented [3, 4] as a computationally simple alternative.
PPC is designed for coherently modulated communication from a point in the water column to an array with
vertical spatial diversity. In its most simple form, the
communicating source sends a probe signal followed by a
blanking period to allow for channel clearing followed by
the communication symbol stream. The receivers initally
see the probe signal followed by the time-spread multipaths. Each channel uses this estimated channel impulse
response (different for each receiver) as a matched filter
for the following communications symbols. The resulting
matched filter outputs are integrated over the array. For
a vertical array that spans the water column with dense
enough spatial sampling, the result can be shown to cancel nearly perfectly inter-symbol interference [4], analogous to the theoretical foundations of the active phase
conjugation phenomenon. Time variability of the channel

impulse responses can be measured by re-probing, which

does impose a certain data rate overhead.

EXPERIMENTAL RESULTS
An experiment to demonstrate the effectiveness of
PPC for underwater acoustic communication was performed in Puget Sound near Seattle, WA in May 2000.
Some results of that experiment were previously presented [3, 4]. Those results were for cases where the receive array spanned nearly the entire water column. A
similar case is shown in the first example here. For this
part of the experiment, both vessels (source and receiver
deployment) were moored in 30 m of water roughly
650 m away from each other. The source was deployed at
a depth of 15 m.
Figure 1 shows the channel impulse responses as measured on the 14-element receiving array, with element
spacing of 1
7 m. This configuration spanned roughly
74% of the water column. The probe was sent, followed
by a 50 ms blank window to measure the channel impulse response, subsequently followed by the communications symbol stream which can be seen at the right side
of Fig. 1. The average estimated SNR per channel for this
case is 12.9 dB using the equation


14

N50

N50

n 1

i 1

j 1

SNRest

xi  n 2 


14

N50

n 1

j 1

x j  n2 

x j n

2

(1)


where N50 is the number of data samples in a 50 ms window, the indices i correspond to samples after the start
of the probe reception, the indices j correspond to samples before the probe reception (i.e., noise only), and the
x i  j
 n are the real-valued data samples from receiver n.
The data were transmitted as a random (but known)
sequence of binary phase shift keyed (BPSK) symbols at

SESSIONS

Output of DBPSK Demodulation: 130163949

Raw Hydrophone Data: 130145916

1.5

1
2
1

0.5

5
Constellation

Receiving Hydrophone

6
7
8
9

0.5

10
11
1

12
13

1.5
0

14
1.04

1.05

1.06
1.07
Time (s)

1.08

1.09

Time (s)

FIGURE 1. Probe pulse response on 14-element array with

start of data sequence, element spacing  1 7 m.

FIGURE 3. Differential BPSK demodulation results, element

spacing  1 0 m.

Output of DBPSK Processing: 130145916

1.5

was still achieved. Note, however, that the constellations

have widened and moved closer to 0, meaning a higher
probability of symbol error.
Future work on this topic will include investigation
of the algorithms performance for non-vertical arrays.
We will also look at the computational and performance
trade-offs involved with using decision directed block estimation of the channel impulse response in place of the
isolated probe concept.

Constellation

0.5

0.5

1.5
0

ACKNOWLEDGMENTS
1

Time (s)

FIGURE 2. Differential BPSK demodulation results, element

spacing  1 7 m.

This work was funded by the Office of Naval Research.

REFERENCES
a rate of 2174 symbols per second, and demodulated differentially (i.e., phase shift between symbols, rather than
absolute phase, is important). The result of the demodulation can be seen in Fig. 2. Ideally, values differentially
encoded as 1 will take on a value of 1.0 in the constellation plot, and values encoded as 0 will take on a value of
-1.0 in the plot. The x symbols denote symbols encoded
as 1, and the o symbols denote those encoded as 0.
Note that there are no symbol errors in this data set.
Roughly 100 minutes later, data was taken with a
re-deployed receive array whose element spacing was
1
0 m, i.e., spanning roughly 43% of the water column. Figure 3 shows the demodulation results for this
case. At this time the noise background had increased,
with SNRest  2
2 dB. Despite the decreased SNR and reduced water column coverage, error-free communication

1. W. A. Kuperman, W. S. Hodgkiss, H. C. Song, T. Akal, C.

Ferla, and D. R. Jackson, J. Acoust. Soc. Am., 103, pp. 2540.
2. D. R. Jackson and D. R. Dowling, J. Acoust. Soc. Am., 89,
pp. 171-81.
3. D. R. Jackson, D. Rouseff, W. L. J. Fox, C. D. Jones, J.
A. Ritcey, and D. R. Dowling, J. Acoust. Soc. Am., 108,
p. 2607.
4. D. R. Jackson, D. Rouseff, W. L. J. Fox, C. D. Jones, J. A.
Ritcey, and D. R. Dowling, "Underwater acoustic communication by passive phase conjugation: Theory and experimental results," IEEE J. Oceanic Eng. (accepted for publication).

SESSIONS

Optimal Superimposing of the Normalized Quasi Rigid

Echo Spectrum on the Quasi Rigid Form Function :
Targets Sizing
P. Schweitzera, J. Mathieua,b, E. Tisseranda, D. Bellefleurb
a

L.I.E.N., Facult des Sciences, Universit Henri Poincar - NANCY I, B.P. 239, 54506 VANDOEUVRE-LESb
NANCY CEDEX, FRANCE.
G.E.M.C.E.A. - 149, rue Gabriel Pri - 54500 VANDOEUVRE-LES-NANCY
We developped a methodology based on the analysis of the backscattered echo from an immersed target for the
determination of its size. The Quasi Rigid Backscattered Echo (QRBE) defined as the first part of the backscattered echo
contains the size information of the target. The Quasi Rigid Form Function (QRFF) is then constructed by considering
only the quasi rigid backscattered echo. The Normalized QRBE spectrum obtained by practice can be interpreted as a
segment of the QRFF. The object of this paper is to present how we find the optimal super imposition between them in
order to find the size of the target. We present successively the two steps for the method. First we determine the classes of
solutions according to the system bandwith B in use. These classes are defined by all the size for which the QRFF has the
same and type of extrema. We improve then this estimate by mean squares criterian applied in this class of belonging. We
present the experimental setup and show different size determination results for different steel wires.

INTRODUCTION
We propose an ultrasonic method to determine the
size of an immersed wire. This can be done comparing
its quasi rigid backscattered echo spectrum on a
reference function. The different steps of the method
of surimposition in order to find the size of the target
are presented successively in this paper.

THE QUASI RIGID BACKSCATTERED

ECHO (QRBE)
We define the QRBE, for an immersed target
(wire) in water and insonified by a plane ultrasonic
wave, as the first part of the backscattered echo. Its
duration is equal to the backscattered echo of an
infinitively rigid wire of same size.
The first part of the echo is mainly constituted by
Scholte-Stoneley and specular waves which contain
the size information of the target [1].

THE QUASI RIGID FORM FUNCTION

(QRFF)
Contrary to the form function which expresses all
the scattering waves generated by a wire, there is no
analytical expression of the QRFF fQR(ka). The quasi
rigid form function is calculated by taking into account
the quasi rigid part of the echo we simulate. The
details of the method is given in [2].

Relative amplitude

1
1
1
0.8
0

ka

FIGURE 1. Modulus of the steel QRFF versus ka.

EXPERIMENTAL CONDITIONS
The samples (steel wires) are vertically imbedded
in a tank filled with still water. The emitter-receiver is
a broadband transducer of 1 cm diameter, acting in
frequencies in the range of 0,7 MHz to 1,5 MHz.

SIZING BY SPECTRAL ANALYSIS.

Method
The first step of our method is the calibration.
block

We calculate the spectrum EQR (f) of the QRBE from

a block which simulate a solid half space. Then this
block is replaced by the wire to be sized
wire

The ratio E NQR (f) =

EQR (f)
block

can be interpreted as

EQR (f)

a segment of the quasi-rigid form function.

SESSIONS

square algorithm in order to find the target size.

wire

EQR (f) is the quasi-rigid part of the echo.

This relation is only valid in the limited bandwidth
B of the measuring chain so for ka limited to the
f
f

interval 2p min a ; 2p max a . c is the sound speed

c
c

in water and a the size of the target

The last step is to determine the size of the wire
under test by replacing the segment E NQR (f) on the

PERFORMANCES
In order to test the performances of our method, we
have superimposed the segment

wire

EQR (f)

on the

QRFF. After identification, we obtain a=394 mm, the

real radius is a = 400 10 .

QRFF.
1.15

Positioning the segment on the QRFF

The first step consists in dividing the QRFF in
several parts to group together different target sizes in
the same class. For a given class, corresponding to a
group of sizes, the segment of QRFF in the bandwidth
of the system will have the same number and type of
extrema.
2
Relative amplitude

0.5

1.5
2
Frequency (MHz)

2.5

FIGURE 2. FFQR for the extreme radius of a class of size

Between these two limit values amin and amax , both

QRFF present two extrema (figure 2).
In a second step, the analysis of the practicle segment
E NQR (f) allows to place it in his own class.
Figure 3 show the normalized spectrum for a steel
wire of 1 mm diameter. The segment has 2 extrema.
Relative
amplitude

0.5

fmes( n )
0.65
80

140

FIGURE 4 : Optimal super imposing of ENQR on FFQR for

a steel wire of radius 400 mm.

CONCLUSION

amin
amax

f( n )

A new concept to find the size of a immersed target

(wire) using the first part of the backscattered echo is
presented in this paper. The methodology is to super
impose the spectra of the experimental quasi rigid part
of the echo on a reference function. The
measurements, performed with steel wires immersed
in water, permit to determine their size with a good
precision.
Research work to find the size of sphere target and
wires of different matter (copper, aluminium ) are
also made.

ACKNOWLEDMENTS
This work is supported by the French Lorraine
Region and G.E.M.C.E.A.

REFERENCES
Reduced
Bandwidth B
[0.75 ; 1.45]

0.75
1.45
Frequency (MHz)
FIGURE 3. Normalized spectrum of steel wire

The last step is to minimize the euclidian distance

between E NQR (f) and the FFQR with an mean

1. H. Uberhall, Surface waves in acoustics, Physical

Acoustics vol.10, Academic Press, 1973.
2. Mathieu J., Schweitzer P., Tisserand E., Bellefleur
D. 10-13 july 2000. Particle size measurements by
spectral analysis : quasi-rigid form function study.
Proceedings of the fifth European Conference on
Underwater Acoustics, ECUA 2000, Lyon (France),
pp. 1887-1892.

SESSIONS

Implosion Sound Sources

N. Yen
CLY Associates, P. O. Box 6806, Alexandria, VA, 22306-6806 USA
Acoustic pressure caused by implosion has impulsive characteristics similar to those generated by an explosive, but the signature is
highly temporally and spatially localized without the safety issues with explosives. The present study focuses on the practical utilization
of a highly concentrated pressure pulse caused by an implosion mechanism as a sound source. The operational principle of an implosion
sound is simply derived from the conversion of hydrodynamic energy to acoustic energy from the collapse of a cavity. However, the
controllability of an implosion can be manipulated by directing the one-dimensional flow into a confined specially shaped cavity to form
the spherical convergence. Based on the physics of the collapse of a spherical cavity, the reflection of a self-focused wave can create a
delta-function like pressure pulse. For underwater application, because no air is trapped in the closed space initially, the acoustic
signature is a simple pressure pulse without any interference of bubbles generally observed in the explosive or air gun operations. The
simple structure of the implosion mechanism can be adapted for various engineering designs to meet the practical needs in
oceanographic and underwater research activities.

INTRODUCTION
Current technology for generating impulsive sound waves
for underwater surveying and communication applications
employs the detonation mechanism. The acoustic signal
obtained in this way is not controllable: it consists of
interference caused by bubbles' resonances and it is
sensitive to the variation of charges and operational depth.
Other methods use various types of frequency synthesis
approaches to form a broad band signal. However, such a
sound source is not temporally and spatially localized, and
its efficiency is low due to the compensation of frequency
response during the electro-mechanical transduction. A
highly concentrated pressure pulse caused by an implosion
mechanism appears to alleviate those problems. A brief
review on some of the implosion underwater sound sources
is given in references 1. The presentation covered in this
short paper is focused on the control of the implosion sound
generation. Extensive discussion of this subject is described
in references 2 and 3.

(e) tremendous temperature at the center causes

dissociation, (f) light emits from black body radiation or
ionization, (g) the reflected implosion propagates out as a
sound pulse. The acoustic emission from a 65 mm bubble
measured in the laboratory [4] has an intensity magnitude
about 170 dB/1mPcal.
The estimated conversion
efficiency from the mechanical energy to the acoustic
energy is about 75%.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

FIGURE 1. Phenomenological Interpretation of Implosion

IMPLOSION MECHANISM

CONTROLLED IMPLOSION

Implosion is formed by directing waves toward a focal

center. Because of the energy concentration, an extremely
strong pressure and high temperature can be built up in a
very small region, thereby generating a sharp shock wave
propagating outwards. A simple mathematical model of this
phenomenon can be
formulated by the collapse of a spherical void such as the
Rayleigh-Plesset's equation used in analyzing the
cavitation. The phenomenological interpretation of the
implosion process sequence shown in Fig. 1 are: (a) the
collapse of a bubble generates a spherical shock wave, (b)
the implosion shock wave converges to the cavity center,
(c) a high pressure pulse is created by energy concentration,
(d) the intensity of a shock front is enhanced by reflection,

The practical way to generate a controlled implosion is

through the manipulation of a plane shock wave [5]
according to the Chester-Chisnell-Whitam (CCW) model
[6] by forcing it to converge to a single point. Fig. 2(a)
shows closed tubes with three different shaped terminals:
(i) cylindrical, (ii) CCW form, (iii) conical. The measured
waveforms of implosion sound from those controlled
enclosed cavities of one inch diameter tube are displayed
in Fig. 2(b) respectively. The highest peak pressure pulse
can reach to an intensity of 190 dB/1mPcal. The CCW
shaped terminal provides a well defined pulse with a
duration of less than 100 ms.

SESSIONS

FIGURE 3. An Implosion Sound Projector

DISCUSSION

(a)

.15

.3 ms

(b)
FIGURE 2. Enclosed Cavity Implosion

IMPLOSION SOUND PROJECTORS

Various types of implosion sound projectors [3] can be
constructed based on the controllable implosion scheme
described in the previous section. Their design parameters
depend on the application requirements and cost
constraints. The sketch shown in Fig. 3 is a project element
which has the feature that can be integrated with undersea
survey system for directional scanning. This sound
generated unit consists of a housing #61, the implosion
chamber # 14 with its implosion center at #16. An
electronic controlled valve #66 is used to create an artificial
void in #14 with a vacuum pump. For the implosion
operation, the valve switches the tube connection to a highpressure tank and forces the fluid #38 inside the housing
squash to the tip of the chamber #62 and causes a
convergent collapse center at # 16. The cover #57 is an
acoustic transparent material so the generated acoustic
wave can be directed out through this window. Many of
such a unit #40 can be arranged to form an array. With the
proper placement of implosion units and timing delay
opening of control valves, a directional radiated sharp
implosion acoustic impulse can be formed. Preliminary
tests of a chamber size of 1" diameter and 1.5" in depth in
the laboratory tank operated with vacuum of 0.022 ATM
and a high-pressure of 1 ATM, generates an impulse of
peak intensity 195 dB/1mPcal with a pulse width less than
50 ms at 3 feet away from the implosion projector can be
directed to a desired direction for remote scanning search.

The use of an implosion sound source described here

has demonstrated that an underwater acoustic wave
generator designed with this type of mechanism is simple
and safe to operate for undersea remote surveying
application. At the current stage, only a simple prototype
has been constructed and laboratory tests have
demonstrated its practicability for underwater application
as an effective sound source. Because of the simplicity of
the implosion mechanism in comparison with the
traditional acoustics radiators, this type of a sound source
has much better performance features in the areas of
acoustic signature (a temporal and spatial concentrated
wide band signal), operation handling (light weight),
system integration (small size), and low cost (less logistic
support). Many alternative designs other than those
mentioned in this paper can be adapted for some
engineering modifications to meet the needs in
oceanographic and underwater research activities. Future
effort for implosion acoustics studies will be directed to
the optimization of the structure design based on a
specific task requirement.

REFERENCES
1.

Yen, N., "Application of Implosion to Underwater Sound

Generation, J. Acoust. Soc. Am. 100, p 2716, (1996).

2.

Yen, N., An Implosion Sound Source for Undersea Exploration

Applications, in Proceedings of Second International Ocean and
Atmosphere Conference, Central Weather Bureau, Taipei, 2000,
pp. 315-318.

3.

Yen, N., An unpublished NRL proposal and "Controllable

Implosive Sound Projector", Reg. Number H1664, United States
Statutory Invention Registration, July 1, 1997.

4.

Stottlemyer, T.R., An Experimental Study of the Acoustic Emission

from Collapsing Cavities in Liquids, A Ph. D. dissertation, Yale
University, New Haven, 1996.

5.

Yen, N., "Controlled Implosion Sound Generation", Nonlinear

Acoustics in Prospective, 14th International Symposium on
Nonlinear Acoustics, Ed. R. J. Wei, Nanjing University Press,
Nanjing, 1996, pp 292-297.

6.

Whitham, G.B., Linear and Nonlinear Waves, Wiley-Interscience,

New York, 1974

SESSIONS

Surface Acoustic Waves for Sediment Characterization;

from Sonar to Tomographic Approach
M. E. Zakharia a and E. Mouton b
a

French Naval Academy IRENav, 29240, Brest Naval, France, zakharia@ecole-navale.fr

b
SAGE-GEODIA, Les Clachs, 34560, Poussan, France, emo@sage-geodia.com

Surface acoustic waves (such as Stoneley-Scholte Waves (SSW) travel along the interface between the seawater and the
seabed; they are guided in a layer of about a wavelength and carry information on the first meters of sediment. Their velocity
can be used for the inversion of fine properties of the sediment. Several experiments have shown that these waves can also be
used for the detection of buried objects. Experiments on tomographic reconstruction of an anomaly in the sediment using SSW
(impedance changes due to the presence of bubbles, for instance) have provided very interesting results and highlighted the
relevance of such a technique for sediment description at sea.

INTRODUCTION
Several applications (such as offshore, cable and
pipeline installation, mine detection, propagation
prediction, slope stability studies) need an accurate
knowledge of the seabed properties. This work will
show how Stoneley-Scholte Waves (SSW) can be a
relevant tool for seabed characterization and subbottom imaging.

Fine characterization of sediments

As the penetration depth of SSW depends on the
frequency, each frequency bin carries information on a
corresponding layer; when using wideband signals, the
group velocity dispersion of the SSW depends on the
velocity profile in the sediment. The group velocity
can be easily measured on a time-frequency
representation of wideband transmitted signals [6].
Several simulations and experiments have shown that
velocity dispersion could be predicted with accuracy
better than 5% (direct problem) [6].
The inverse problem consists in determining the
sediment properties from SSW properties. A new
approach, based on neural network has been developed
for this purpose. It showed, on experimental data, that
a comparable accuracy (better than 5%) can be
obtained on the data from inverse problem solution
[5]. Such results were very encouraging: if SSW
properties are sensitive to small changes in the
sediment, they should be very sensitive to the presence
of a buried target (large variation of impedance).

Detection of a buried object

Several tank experiments have been carried out on
buried targets in a homogeneous resin [4]. For a
transmitter position, the surface was finely scanned
and the SSW energy on the interface was computed.
Figure 1 displays the results of such a computation. In
this figure, one can see from top to bottom:
the incoming wave loosing some energy while
propagating,
a zone of interference between the incoming wave
and the one reflected by the sphere
a shadow zone after the sphere very similar to
shadows encountered in sidescan sonar.

FIGURE 1. Energy distribution of SSW at water

bottom interface in the presence of a spherical target
size: 16 mm; frequency 0.1 MHz, wavelength 10 mm.
2-dB/ gray level, scales 0.1 x 0.1 meters.
Similar effects have been observed for various targets
(even when comparable in size to a wavelength). They
showed that the presence of a target scatters the energy
of SSW in all direction (instead of forward direction).

SESSIONS

The absence of energy can thus be used to detect the

presence of a target (like in sidescan sonar).
Preliminary work (still in progress) has shown that a
sonar approach (monostatic transmitter and receiver)
can also be used for target detection.

Reflection and transmission properties

The reflection and transmission of SSW was studied in
the solid-solid configuration. As they are evanescent
waves, two hypotheses have been found: continuity
conditions for either each component or their resultant.
From experiments, we have found that the second one
matches better and that, at oblique incidence, SSW
follow laws similar to Snell-Descartes ones [2]:
SSW is reflected as a SSW (with same velocity)
SSW is transmitted as another SSW (with a SS
velocity corresponding to the second medium)
A critical angle was observed (similar to the one
encountered for compression waves).
No other waves or components were observed.

FIGURE 2. Tomographic reconstruction of a

cylindrical inclusion in a resin.
Scales: 0.2 x0.2 m, 18 m/s by gray level

REFERENCES
1.

M. E. Zakharia and P. Chevret Neural network

approach for inverting velocity dispersion; application
to sediment and to sonar target characterization, Inverse
Problems 16 (2000) 1963-1708.

Tomography

2.

The properties cited above allowed a tomographic

reconstruction of velocity using SSW. Several mockups and geometrical configurations were studied using
a cylindrical inclusion in homogeneous sediment [3].
Figure 2 shows an example of tomographic
reconstruction of the SSW velocity. The position of
the inclusion has also been displayed on the figure.
The results clearly show the high quality of
tomographic reconstruction using the backpropagation
method.

E. Mouton, J. Chtillon et M.E. Zakharia, Etude

exprimentale de la rflexion et de la transmission des
ondes de surface de type Stoneley-Scholte l'interface
de deux milieux solides, Actes du 5e Congrs Franais
d'Acoustique, CFA 2000, septembre 2000, Lausanne,
Suisse, pp. 80-83.

3.

E. Mouton and M. E. Zakharia, Reconstruction of

sediment inhomogeneities using surface wave
tomography, in Proceeding European Conference on
Underwater Acoustics (ECUA2000), Lyon, July 2000,
M.E. Zakharia, P. Chevret and P. Dubail editors,
European Commission Brussels (Belgium), Vol. 1 pp.
245-250.

CONCLUSION

4.

M.E. Zakharia and J. Chtillon, Interaction of interface

waves with a buried object, in Proceedings of The Third
European Conference on Underwater Acoustics
(ECUA), Heraklio (Greece), June 1996, J.S. Papadakis
Ed., European Commission Brussels, pp. 39-44.

5.

J. Guilbot and M. Magand, Determination of the

geoacoustical parameters of a sedimentary layer from
surface acoustic waves: a neural network approach,
Conference on Full Field Inversion Methods in Ocean
and Seismic Acoustics, O. Diachock, A. Caiti, P.
Gerstoflt and H. Scmidt Eds, Kluwer Academic
Publishers, pp.171-176, 1995.

6.

J. Guilbot and M.E. Zakharia, Tank experiments on a

sediment small-scale model. Shear wave velocity
profile inversion via Stoneley-Scholte waves, in
Proceedings of The Second European Conference on
Underwater Acoustics (ECUA), Lyngby (Denmark),
July 1994, L. Bjrn Ed., European Commission,
vol. II, pp. 979-984.

Results from various tank experiments and associated

signal processing schemes showed the ability of SSW
to provide accurate information on fine characteristics
of the sediments and to be used for sub-bottom
imaging and the detection of buried objects. Next step
is the application of the techniques described and
validated in tank to the seabed and to real applications
(in situ).

ACKNOWLEDGMENTS
The work was achieved in LASSSO laboratory
(Laboratoire dAcoustique, Systmes, Signaux et
Sonar, at CPE, Lyon.) and was partly supported by the
European Commission and by the French MOD.

SESSIONS

Stationary hydroacoustic methodology to determine diel

activity of fish biomass in the artificial habitats
Sala A.(1), Fabi G. (1)
(1)

Istituto di Ricerche sulla Pesca Marittima (IRPEM), Consiglio Nazionale delle Ricerche
Largo Fiera della Pesca, 1 60125 Ancona, Italy

The biomass of fish assemblage, inhabiting the Senigallia artificial reef (central Adriatic sea, Italy), was evaluated in the period
JulyNovember 1996. Density and biomass were assessed through a stationary hydroacoustic methodology using an
appropriately adapted SIMRAD EY500 system. A part of the system was placed inside the reef and it was linked by radiomodem to the remaining part installed ashore in the Institute. The experimentation gave useful information about the daily
behaviour of the fish assemblage living at the reef: during the whole period the lowest densities were generally recorded in the
early afternoon, whilst the highest abundances were commonly observed late in the night and in the early morning. Acoustical
records confirmed that in late summerearly autumn most of the reef fishes migrate from the coastal shallow waters to offshore.
Throughout the study period the fish abundance was higher inside the reef and decreased significantly at a distance of about 80 m
from the structures.

An acoustic fixed technique was applied for the

valuation of the fish biomass at an artificial reef
deployed along the coast of the central Adriatic sea
(1.2 nmi offshore, 12 m depth).
The hydroacoustic equipment consisted of Simrad
EY500 echo-sounder and comprised:
- 1 transceiver with Personal Computer;
- 4 batteries (charged with solar panels);
- 1 transducer-multiplexing (SIMRAD MP500);
- 1 timer, to periodically power the system.
All the above equipment, was placed inside a
waterproof case on a fixed buoy inside the reef area.
The EY500 system was linked by a radio-modem to
another Personal Computer installed ashore in the
Institute, which through an appropriately developed
program automatically controlled the correct
functioning of the EY500 system in real time.
Four split-beam transducers (120 kHz) were settled
to measure in situ fish Target Strength distribution and
density:
- transducer 1 (T1), placed 4m-deep on the fixed
buoys pile, was horizontally oriented towards the
centre of the reef;
- transducers 2 and 3 (T2, T3) were located on steel
frames placed on the bottom inside the artificial reef
and upward-oriented;
- transducer 4 (T4) was oriented towards the surface,
as T2 and T3, but located on the bottom outside the
artificial reef, about 80 m far.
Every two hours the system was powered on by the
timer for a period of 16 minutes and the echo-sounder
started pinging immediately.

The equipment operated continuously and the

acoustic data, received from the echo-sounder, were
stored on the computers hard-disk in a telegram-based
structure. Contemporarily, every 60 seconds, the
system transferred through the radio-modem the data
integration to the Institute, to allow the control of data
acquisition.
The whole study period was subdivided into 8
intervals, each having a duration of about eight days
(Table 1). During each interval the system operated
continuously 24 h/day until the hard disk saturation;
afterwards the system was reset.
Table 1. Study period (21 Jul 96 14 Nov 96). Duration of
the 8 sampling intervals
1
21/7
Start

2
2/8

3
4
5
11/8 20/8 2/9

6
25/9

30/7

8/8

17/8 29/8 9/9

30/9 12/10 14/11

End

7
2/10

8
1/11

50

T1

T2

T3

T4

40

gr/m

EXPERIMENTAL DESIGN

30

20
10
0
1

Sampling interval

Figure 1. Mean fish biomass recorded by the four transducers

during the whole sampling period

SESSIONS

Density

NV (%)

100%

Biomass

100%

80%

80%

60%

60%

40%

40%

20%

20%

NV (%)

T1

T2

0%

0%

100%

100%

80%

80%

60%

60%

40%

40%

20%

20%

T3
0%

T4

0%
0

10

12

14

16

18

20

22

Hour of the day

10

12

14

16

18

20

22

Hour of the day

Figure 2. Mean fish density and biomass recorded by the four transducers during different hours of the day. For each
transducer the mean fish abundance (density and biomass) at each hour were normalized dividing by the maximum meanvalue (NV%)

RESULTS AND CONCLUSIONS

The present study demonstrated the suitability of the
fixed hydroacoustic techniques to get ecological and
practical information on the fish assemblage living
inside and around artificial structures. The values
recorded by the off-reef transducer (T4) were generally
lower than those collected by the other ones (Figure 1),
evidencing that the reef effect on the fish assemblage
was already reduced at about 80 m from the structures.
This agrees with the results of other researches
indicating that the local area of influence of an
artificial reef may range from 5-50m, depending on the
local environmental conditions and on the reef size [2].
Moreover, the fish abundance did not appear
homogeneously distributed inside the reef: the highest
densities were recorded in the central part of the area
(T2), where there is a higher concentration of
structures. In harmony with a previous study [1], the
acoustical records also confirmed that in late summer
early autumn most of the reef fish species migrate
from the coastal shallow waters to offshore where,
during the winter months, the water temperature is
about 10-12C. Finally, the experimentation gave
useful information about the daily behaviour of the
fish assemblage living inside the reef (Figure 2). The
current fish biomass measurements within diel period
corroborated the earlier findings [4]: a minimum of

density was generally recorded during the early

afternoon, while the highest abundance were
commonly observed late in the evening, during the
night and early in the morning. A recent study [3] gave
similar results on diel acoustic measurements, but the
author correlated the 24-h acoustic fluctuations with
the hydrographic factors such as temperature, oxygen
level and salinity, that have important influence on fish
physiological state. Because Senigallia reef area has
relative stable hydrography, the associated effects of
the previous factors have probably little influenced the
conversion of acoustic data into fish abundance.

ACKNOWLEDGEMENT
The authors are in debt to Dr. Loris Fiorentini
(IRPEM-CNR Ancona) for his huge effort in the
experimental design, set-up of the echo-sounder
system and his support in the field work.

REFERENCES
1. G. Fabi and L. Fiorentini, Bull. Mar. Sci. 55, 538-558
(1994).
2. F. Gerlotto, C. Bercy and B. Bordeau, Proc. Inst. Acoust.
19, 79-88 (1989).
3. A. Orlowski, ICES J. Mar. Sci. 57, 1196-1203 (2000).
4. R.E. Thorne, J.B. Hedgepeth and J.A. Campos, Rapp. P.v. Run. Cons. Int. Explor. Mer. 189, 167-175 (1990).

SESSIONS

Using Adaptive Algorithms in Indirect Fish

Target Strength Estimation
M. Moszynski
Technical University of Gdansk, Department of Remote Monitoring Systems,
ul. Narutowicza 11/12, 80-952 Gdansk
The typical approach to the problem of indirect fish target strength estimation from data collected using single-beam system is
based on transforming probability density function (PDF) of measured fish echo level into fish target strength PDF estimate. In
transformation algorithms, the PDF of beam pattern, which represents the kernel of transform, has to be known. Furthermore, the
calculation of beam pattern PDF depends on assumed distribution (typically the uniform) of fish in the water column. However,
due to ill-conditioning of most transformation algorithms, small errors in input data, and inaccuracies in the assumed form of the
kernel, may result in large errors at the output. Therefore, most of modern inversion algorithms use sophisticated techniques to
achieve satisfying results, but assuming fixed kernel in the inversion procedure.. The paper presents different approach, which
uses adaptive construction of the kernel. As a result, the optimal beam pattern PDF is obtained which leads to more reliable
estimate of fish target strength PDF, than in fixed kernel methods.

INTRODUCTION
Indirect fish target strength estimation when using
single-beam echosounder data leads to the inverse
problem in which the probability density function
(PDF) of target strength is estimated from fish
echoes. Mathematically the problem is described by
so-called single-beam integral equation , as a
convolution-like integral of the following form [1]:

p E (E) =

Bmin

p B ( B ) pTS ( E - B ) dB

(1)

where E represents echo level (E=TS+B) and Bmin is

the lower threshold of logarithmic beam pattern
function included in calculations.
Due to the hydroacoustic system characteristics
the reconstruction is based on incomplete data. This
kind of problem is an example of a statistical linear
inverse problem (SLIP), often presented as a linear
operator equation y=K x , where observation y is
represented by echo level peak values PDF pE(E),
linear operator K (kernel) is constructed from
logarithmic beam pattern PDF pB(B) and x is
unknown function representing fish target strength
PDF pTS(TS). Statistical linear inverse problems are
typically ill-conditioned and can be solved using
direct inverse techniques based on regularization (i.e.
Windowed Singular Value Decomposition - WSVD)
[4] or iterative ones in which additional constraints
are specified (i.e. Expectation, Maximization,
Smoothing - EMS) [3].

INVERSE METHODS
A number of references to the earlier work on
indirect target strength can be found in [2]. In [3] and

[5] the authors investigated some of the earlier methods

and introduced some novel inverse techniques.
Generally, two kinds of inverse methods direct and
indirect (iterative), described below, are used.
Direct inverse techniques using regularization are
based on pseudo-inversion in which Moore-Penrose
matrix K# derived from the kernel K is used. This matrix
provides the minimum-norm least squares solution to the
problem of finding the unknown vector x, that
simultaneously minimizes the equation error ||Kxy||2.
This pseudo-inverse matrix can be effectively computed
using SVD techniques and some other modification
applied by introducing weighting factors wj to singular
values gj, leads to solutions in the form [5]:
(2)
xWSVD = w j g -j 1 [ y , h j ] e j
j

gj2

and ej are, respectively, the eigenvectors and

where
eigenfunctions of K*K, normalized image is defined by
h=K/||K||, and [.,.] is the standard inner product in L2
space.
The EMS technique is an example of indirect inverse
technique. The method constrains estimates to be
positive and reduces the time needed to converge by
smoothing groups of estimates per iteration. Every
iteration procedure performed during solution consists of
three
steps
called
respectively:
expectation,
maximization and smoothing. Assuming that observation
y results from a Poisson process we received equation
describing first two EM steps in a form [5]:

x( n -1 )
yt
n)
(3)
x(EMS
( n -1 ) T K
=
K
K
x
i ij

SVD technique gives the solution with minimum

squared error, which is typically used as a natural
measure of global goodness-of-fit test for an estimate.
However, due to sine-like nature of eigenfunctions ej of

SESSIONS

Lake, Idaho survey [4] were used. Over 10000 echoes

were acquired by a dual-beam system operating on
420kHz and post-processed by the sounder software.
Narrow beam data were used for indirect estimation.
Data from both beams were used to construct the
estimate only for comparison purposes. Fig.2a. shows
pTS estimate obtained after EMS step, its verification in
the form of actual pE and pE obtained by convolution of
pTS estimate with assumed pB estimate is presented in
Fig. 2b. Fig. 2c shows reconstruction of beam pattern
PDF pB from the actual pE and pTS estimated just before.
Fig. 2d presents next two estimates of pTS obtained in
successive adaptive steps. Table 1 reports the value of
root-mean-square error for WSVD and three first
adaptive EMS steps. Application of AEMS reduces rms
error and simultaneously represents good estimate for
class of probability density functions. Additionally, as a
result of kernel modifications, more adequate beam
pattern PDF is obtained which leads to more reliable
estimate of fish target strength PDF, than in
conventional methods based on heuristic approach.

linear operator K, SVD often leads to the artifacts

when interpreting obtained estimate as a probability
density function. The EMS estimate represents more
smooth class of functions than those obtained by
SVD and can be treated as a good estimate for a class
of probability density functions, although resulting
mean square error is much larger. This error results
from inappropriate estimate of kernel K and can be
minimized preserving smoothness of solution by the
changes introduced in the kernel K of integral
equation Eq.(1). This is particularly relevant for the
case of fish target strength estimation as the
construction of the kernel is based on heuristic
assumption made on angular distribution of fish in
calculation of beam pattern PDF. Thus, the
estimation algorithm may adaptively change kernel K
by solving another inverse problem in which the
beam pattern PDF pB is reconstructed from echo level
PDF pE and target strength PDF pTS estimated just
before. New estimate of pB PDF allows calculating
new kernel matrix K, which is used in the next step of
such adaptive algorithm. The process can be
terminated comparing the difference between two
successive estimates.

Table 1. Root-mean-square error of WSVD estimate and

successive adaptive EMS (AEMS) estimates.
WSVD

RESULTS

RMS 0.0231
error

To verify the idea of adaptive EMS technique the

data provided by Parkinson from Coeur dAlene
a)

b)

pTS(TS)

c)

pE(E)

EMS

AEMS
(n=2)

AEMS
(n=3)

0.1288 0.0477 0.0426 0.0419

d)

pB(B)

pTS(TS)

1000

700

1000

AEMS
(n=1)

0.14
600
800

800

0.12
500

600

0.1

400

0.08

300

400

400

200

0.06

100

0.04

200

0
-70

-60

-50

-40

-30

0
-20

600

-10

10

20

200

0.02
-40

-30

-20

-10

0
-70

-60

-50

-40

-30

FIGURE 2. a) First EMS reconstruction of the target strength PDF compared with estimate obtained from dual-beam data (thin
line), b) verification of first EMS reconstruction with actual echo PDF (thin line), c) reconstruction of beam pattern PDF
compared with assumed one (thin line) d) two successive adaptive EMS estimates (thin line dual-beam estimate).

REFERENCES
1.
2.

3.

Clay, C.S., Deconvolution of the fish scattering PDF

from the echo PDF for a single transducer sonar. J.
Acous. Soc. Am., 73: 1989-1994.
Ehrenberg, J.E., A review of target strength estimation
techniques. Pp. 161-175 In Y.T. Chan, ed. Underwater
Acoustic Data Processing. Kluwer Academic
Publishers, 1989.
Hedgepeth, J.B., Gallucci, V.F., OSullivan F.,
Thorne, R.E., An expectation maximization and

4.

5.

smoothing approach for indirect acoustic estimation of

fish size and density. ICES J. Mar. Sci., 56: 36-50,
1989.
Parkinson, E.A., Rieman, B.E., Rudstam, L.G., A
comparison of acoustic and trawl methods for
estimating density and age structure in kokanee.
Trans. Am. Fish. Soc., 123:841-854, 1994.
Stepnowski A., Moszynski M., Inverse problem
solution techniques as applied to indirect in situ
estimation of fish target strength, J. Acous. Soc. Am.,
vol. 107, No 5, pp. 2554-2562, Fig. 11, Ref. 28.

SESSIONS

The Three-frequency Method for Classifying the Species and

assessing the Size of two Euphausiids (Euphausia superba
and Euphausia crystallorophias).
M. Azzali, J. Kalinowski, G. Lanciani, I. Leonori
Institute of Fisheries Research, Research National Council, 60125 Ancona, Italy
In this paper an acoustic method for identifying two euphausiid species and estimating their length is described. The approach is
in fact an outgrowth from both the fluid sphere and Bayes rule methodologies. Some practical results of the method are
presented.

THE PROBLEM
The fundamental problem of ecology in Antarctic is
the conservation biology of krill Euphausia superba. In
the Ross Sea two krill species dominate the biomass E.
superba (E.s.) and E. crystallorophias (E.c.). Therefore
target species (E.s. and E.c.) identification and their
size estimation is the basic problem in krill assessment
by hydroacoustic methods. A three-frequency method
for euphausiids discrimination and size estimation has
been developed. This paper explores applications of
the multi-frequency method using data from three
expeditions to the Ross Sea (1980-90; 1997-98 and
1999-2000).

THE CLASSIFICATION METHOD

The fluid model. Theoretical considerations [1,2,3]
demonstrate that the ratio (rfj/fi) of the Mean Volume
Backscattering Strength measured at two different
frequencies (fj/fi) from non resonant marine organisms
can be used to calculate the spherical radius of their
backscattering cross sections. The mean length L of an
euphausiid with equivalent radius (a) can be calculated
approximating its trunk with an equivalent cylinder [4]
and equating the volume of the scatterer to the volume
of the equivalent sphere: L = 12.11 *(a) in mm.
The hybrid model. The above model assumes a
deterministic dependence of rfj/fi parameters on the
volume of the body of a non resonant animal up to
several centimetres, independently from its species.
This is an evident idealization of the reality.
Differences in the rfj/fi parameters for individuals with
similar size but belonging to different euphausiid
species were observed [5] and may be generated by
differences in the physical parameters, in acoustic
behaviour and in shape. These differences are essential
in species recognition, but for the amounts and

complexity of the acoustic processes that they

generate, it would become extremely complicated or
even impossible to include them in the fluid sphere
model, that is quite effective in the classification of
species per size. This kind of difficulty can be
overcome using both statistical methodology and fluid
sphere model in an "hybrid approach" to species
recognition.
In Ross Sea there are two euphausiid species or
target classes 1=E.s and 2=E.c. We assume that the
set of acoustic samples, taken in each expedition can
be correctly assigned to one of two possible classes on
the basis of net samplings. All the samples that could
be misclassified (mixed hauls, hauls with other
scatters) were attributed to a class 0 and were not
considered in the classification process. Samples
corrupted by noise were discarded. Therefore the set
of the selected measurements s(h) h=1,2, acquired in
each expedition, was partitioned into two independent
sets: s(1) acoustic samples assigned to E.s; s(2)
acoustic samples assigned to E.c.
The measurements of each set can be represented by
random vectors. The three components of a random
vector s v are the outputs from the transducers working
respectively at 38, 120 and 200 kHz.
The range of each component was divided into a
fixed number (n = 40 Log fj/fi) of equal intervals (1
dB). The DMVBS calculated for each pair of frequency
(fj, fi; fj>fi)) are: fj/fi ( fj/fi=10Log(rfj/fi) = 10(Log svfj Log svfi). The number of fj/fi in the bins bm (m=1,2
n) belonging both to the class 1 and to the class 2
define the histogram estimate of the unconditional
p.d.f: p(fj/fi). The three histograms provide a realistic
picture of the dependence of fj/fi on the classes h, of
mutual class overlap, of class separability and of class
probabilistic structure.
The normality test was applied to the conditioned
p.d.f.s. of both classes. The test enabled us to assume

SESSIONS

that the histograms p(fj/fi/h), tend to gaussian

distributions, when the number of observation becomes
large (samples from all the expeditions). We assume
each class can be adequately represented by the three
gaussian or prototype p.d.f.s., estimated from the
relative histograms of correctly classified samples. The
unconditional probability density functions governing
the distributions p(fj/fi), for each pair of frequency fj, fi
(fj>fi) were calculated. Because it is only scarcely
known the probabilistic distribution of E.s. and E.c.
and it can continually change as a result of, perhaps,
geographical location and environmental conditions
we assume that P1=P2= 0.5 (Ph= a priori probabilities
of the classes h; with h=1,2). Using Bayes Theorem,
the a posteriori probabilities p(h/fj/fi) of the classes
h, were found, for each pair of frequency fj, fi (fj>fi).
It was assumed that the classifier assigns to the class
h, each component of a vector generated by a layer m
and belonging to an unknown class x ( m(x) = (200/120;
200/38; 120/38) m(x)), using the Bayesian decision
criterion: decide to assign fj/fi to the class 1 if
p(1/fj/fi) > p(2/fj/fi) or to the class 2 if p(1/fj/fi) <
p(2/fj/fi). Each component of the vector m(x) is
classified independently. The error incurred in
classifying a component fj/fi, using the above criterion
and the divergence (Dfj/fi) between the classes 1 and
2, for each pair of frequency fj, fi (fj>fi), was
calculated to test their separability. The upperbound on
the error E(fj/fi) can be expressed in terms of Dfj/fi
(Table 1). The Bayesian decision criterion classifies
the individual components of a vector m(x)
independently. The final decision rule is to assign one
class to the vector, given the decisions on each of its
components (200/120; 200/38; 120/38)m(x). We used the
"majority vote rule": the class assigned at least to two
components out of three is assumed as the correct class
of the vector. If no majority is got, "no decision" is
taken. Three distributions of the equivalent radii are
obtained from a classified vector. The largest members
of the scattering layer are detected from the couple of
frequency (120, 38 kHz), the mean members from the
couple (200, 38) and the smallest members from the
couple (200, 120). The weighted mean (weight =
(svfj*svfi)0.5) of the three equivalent radii obtained from
the layers sampled by the net were correlated with the
mean length of the relative haul. The obtained
relationship (a regression line for each species) was
used to estimate the value of the length that it occurs
when the mean equivalent radius was calculated. The
classification method was tested using the
"resubstitution error-count estimator". The same sets
s(1) and s(2) of acoustic and biological data, used to
design the method, increased with the set s(0) of
mixed data, was used to estimate the performance of
the method. The three-frequency classification method

was also compared with the "single-frequency method"

in the estimation of the biomass of E.s. and E.c. in the
surveys carried out in December 1997 and in January
2000. In the single-frequency method the size and the
species are deduced from the catches.

RESULTS
The discrimination criteria used to discriminate the
two species for the three pairs of frequencies are
reported in Table 1.
Table 1. Threshold levels in dB
x1
x2
x1
x2
Dfj/fi
E. superba
E.crystallorophias
-0.61
5.23
3.85
8.80
D200/120
6.51
18.84
18.12
27.66
D200/38
6.12
14.08
12.94
20.16
D120/38

E(Dfj/fi)
%
<17
<12
<15

Using the discrimination thresholds shown in table 1

the 91.3% of the 103 E.s. aggregations and the 96.6%
of the 59 E.c. aggregations sampled by the net, were
correctly classified. The correlation between the
biological and acoustical mean length for E.s. data was
highly significant for 1997-98 (Pearson=0.67;
p<0.001) and significant for 2000 (Pearson=0.54;
p<0.05). Also for E.c. the correlation resulted
significant (Pearson=0.53; p<0.05). As an example of
application of the method, the estimations of krill
biomass were made by multi-frequency method. They
differ from -5% (1st echosurvey: from 12 to 17) up to
32% (2nd echosurvey: from 19 to 26) from those
obtained from single-frequency method. Similar results
were obtained from the echosurvey of 2000.

REFERENCES
1.

Johnson R.K. (1977). Sound scattering from a fluid sphere revisited. Journal of the Acoustical Society of America, 61:375377.

2.

Greenlaw C.F., Johnson R.K. (1983). Multiple frequency

acoustical estimation. Biological Oceanography, 2:226-242.

3.

Mitson R.B. Simarad Y. Goss C. (1996). Use of a twofrequency algorithm to determine size and abundance of
plankton in three widely spaced locations. ICE Journal of
Marine Science.

4.

Clay C.S., Medwin H. (1977). Acoustical Oceanography:

Principles and Applications. A Wiley-Interscience Publication:
544 p.

5.

Madureira L.S.P., Everson I., Murphy E.J. (1993b).

Interpretation of acoustic data at two frequencies to
discriminate between Antarctic krill (Euphausia superba Dana)
and other scatterers. Journal of Plankton Research, Vol. 15, no.
7: 787-802.

SESSIONS

Simulation of 3D Seafloor Mapping from Multibeam Sonar

Data Using Electronic Chart Bathymetry Background
M. Moszynskia , Z. Lubniewskia , J. Demkowiczb and A. Stepnowskia
a

Remote Monitoring Systems Department, Technical University of Gdansk, 80-952 Gdansk, Poland
b
C-MAP Poland Ltd., Narutowicza 11/12, 80-952 Gdansk, Poland

The paper investigates 3D mapping of seafloor, which uses modelling of multibeam sonar echoes reflected from seafloor 3D
images, reconstructed from the bathymetry of electronic navigational charts. In the first stage, the 3D relief of seabed surface was
derived from bathymetry soundings data contained in vectorised digital navigational charts. Second stage constitutes the
simulation of the set of hypothetical multibeam sonar echoes scattered on the bottom surface. Finally, the bottom surface is
reconstructed from acoustic data and compared with the images extracted from the charts. The performance of the applied
procedure was evaluated and discussed.

INTRODUCTION
There are known applications of multibeam sonars
in enhanced bathymetry measurements and seafloor
relief mapping etc. The paper presents the simple
procedure of seabed mutibeam echoes modelling along
with application of simulated signals for reconstruction
of bottom relief. The 3D seafloor images used in the
procedure describe real scenes and were reconstructed
from navigational charts.

Also, if possible, the rule of taking the point z only

from this cell, which zi or zj belonged to, was satisfied
during that process.
Such prepared triplets of points constitute 3D
triangulated irregular network (TIN) describing the
bottom surface relief. The example of 3D seafloor relief
image obtained from bathymetry data and processed
using TIN is presented in Fig. 1.

3D SEABED RELIEF
RECONSTRUCTION FROM
NAVIGATIONAL CHARTS
The 3D seafloor images were reconstructed from the
vectorised World Wide Electronic Chart Database CM93. In this process, the Delaunay triangulation method
was used [1].
The input of this procedure was the set of points soundings described by co-ordinates in 3D space. The
first step was to apply the adaptive tree approach to
divide the data points set into cells of varying sizes,
each of which contained no more than m points.
The second step was to obtain the 3D surface by
constructing the Delaunay triangulation. In this step,
the algorithm started from a given point and created the
first edge connecting it with the nearest neighbour.
Then, the successive triangles were created by
assigning the third point z to a given edge zi zj , using
the criterion of minimum distance f(z) from z to zi zj :
f(z) =

( z z ) (z z ) ,
2 (z z ) n
i

r
where n is the unit vector normal to zi zj .

(1)

FIGURE 1. The 3D bottom relief reconstructed from

navigational chart data using trianglated irregular
network (TIN)

SIMULATION AND 3D BOTTOM

RELIEF RECONSTRUCTION
FROM ACOUSTIC DATA
The second stage was to simulate the acoustic
multibeam echoes scattered on 3D relief of seafloor
surface. The set of M N echo signals corresponding
to N beams in each of M scans of multibeam sonar
system over the seafloor surface was generated (see
Fig. 2a). The echoes were simulated using the
following formula for i-th echo waveform ei (t)

SESSIONS

proportional to the signal intensity I, assuming the

domination of incoherent scattering:
ei ( t ) =

e s

0 bs

S i (t )

(inc ) bi ()

R4 ds ,

(2)

where S i (t) - bottom surface insonified by i-th beam, e0

- transmitted signal value, ss (inc) - seafloor
backscattering coefficient for incidence angle inc, b i ()
- i-th beampattern value for transmission angle , R distance to transducer array.
For the seafloor backscattering coefficient angular
dependence, the following formula was used [2]:

ss (inc ) = A exp 2inc + B cos 2inc ,

ct
cti
sini , zi = H i cos i ,
2
2

1
...

(3)

with A, B, and values evaluated using the results of

research shown in [2].
Hypothetical sonar was modelled using parameters
of EM3000 multibeam sonar having operating
frequency 300 kHz and 80 beams with resolution of
1.5. The sonar was assumed to operate 300 meters
above the seabed surface. The distance between
consecutive scans was approx. 6 m.
The next step was to reconstruct the bottom relief
from simulated data. It was performed using the delay
times ti , i = 1, 2, ..., N evaluated for each echo in one
scan. This time was estimated as a point, where the
signal exceeds firstly the 75% of maximum value.
This set of delay times ti was used to reconstruct the
geometric relief of bottom z(x) along the vertical
crossection corresponding to the single scan, by
interpolating (xi , zi ) points, where:
xi =

acoustical simulation algorithm, where a number of

usually important effects was neglected, e.g. the
influence of refraction in water column, the effect of
shading etc.
In the next stage of the investigation, more
advanced acoustic model will be applied, as well as the
experimental validation of proposed methods with use
of actual multibeam sonar data will be carried out.

M
N

1
z
x

a)

(4)

c - sound speed in water, H - bottom depth, i - angle

of i-th beam acoustic axis.
Assuming that the ship moves steadily along the y
axis (Fig. 2a) and consecutive multibeam scans
correspond to successive values yi the relief z = f(x, y)
of the whole investigated seabed surface was
reconstructed by interpolation.

RESULTS AND CONCLUSION

The geometry of the experiment and the
reconstructed 3D bottom from simulated multibeam
echoes are presented in Fig. 2a and b respectively.
It is easy to seen while comparing the pictures from
Fig. 2a and b, that the reconstructed image is quite
consistent with the original one obtained from
soundings. This justifies the practical utility of the
method and confirms adequacy of the algorithm.
However, it must be pointed out that similarity of
obtained images might be also due to simplicity of

b)
FIGURE 2. a) The bottom surface with indicated
geometry of simulated mutlibeam data acquiring
procedure; b) the seafloor 3D image obtained from
sonar data using reconstruction algorithm.

REFERENCES
1. Demkowicz J., Stepnowski A., 3D imaging of seabed
from electronic chart bathymetric data, Hydroacoustics 4,
37-42 (2001).
2. Lurton X. et al., "Shallow water seafloor characterization
for high-frequency multibeam echosounder: image
segmentation
using
angular
backscatter",
in
SACLANTCEN CP-45, La Spezia, 1997, pp. 313-322.
3. Lubniewski Z., Moszynski M., Modelling the seafloor 3D
relief and its reconstruction from multibeam sonar data,
Hydroacoustics 4, 153-156 (2001).

SESSIONS

Jet Noise At High Reynolds Numbers Using Large-Eddy Simulation

and Lighthill's Analogy
D. B. Scheina,b and W. C. Meechama
a

Dept. of Mechanical & Aerospace Engineering, Univ. of California, Los Angeles, California, USA
Northrop Grumman Corporation, Integrated Sys. Sector, Air Combat Systems, El Segundo, California, USA

A computational fluid dynamics model for free, heated jet flow and resultant far-field sound has been developed, which uses largeeddy simulation (LES) and Lighthills acoustic analogy. The procedure involves no adjustable parameters. A deductive, subgrid
scale (SGS) model (based on a Taylor series expansion of the filter function is used for the large eddy simulation. The model can
be run on a Personal Computer, and simulations have been tested using published experimental mean flow field and RMS fluctuation
data for a turbulent, free jets. We have addressed large Reynolds number, high subsonic (compressible) flow with realistic geometries.
In our simulation, Gaussian random velocity fields are introduced at the jet exit to excite the turbulence. The far-field sound and
directivity are computed using the time-derivative form of Lighthills source-integral result which is integrated in time. Simulations
for two power settings of a WR19-4 turbofan engine exhaust (Ma=0.45 and Ma=0.78) were performed, and propagated jet noise
results compared with experimental acoustics data. The agreement is within 2 dB. The experimental agreement shows that the
computed turbulence intensity has an error of but 3%. Other research applications of this approach include the automobile tire noise
due to small jets of air from tread row gaps, background noise in blowdown wind tunnels, and more.

GOVERNING EQUATIONS AND

FILTERING
The
LES
decomposition is
represented
as F = F + F . The filtered variable is defined by the
convolution integral,
F ( x ) = D F i ( z ) G i ( x - z , Di )d z
(1)
D

where G i ( x , D ) is a suitable spatial filter, F can be

termed the large-scale part of F while the residual
portion, F, is the small-scale, or subgrid part. A
Gaussian filter was used in this study,
G( r ) = ( 6 / p D 2 )3/2 e(-6 r
2

/ D2 )

with

( rU~ k )

( rU~ k U~1 )

P s kl t kl
+
+
(2)
t
x l
x k xl xl
where the stress tensor is given by
~
~ ~
t kl r(UQ - U k U l) with U Q = U k U l . . The filtered momentum
equation is solvable (closed) if we provide a model
for t kl . The full system of equations can be found in [2].
+

=-

SUBGRID SCALE TURBULENCE AND

ACOUSTIC MODELING
Smagorinsky [1] was the first to propose a model for
the SGS stresses, assuming that they follow a gradientdiffusion process analogus to heat conduction. His eddy

2 1/ 2

D = (D 1 + D 2 + D 3 )
where i is the filter width in
th
the i direction.
Turbulent flow field equations are derived by
decomposing the dependent variables in the conservation
equations into time mean and fluctuating components.
Here mass-averaged variables are defined according to
~
F = rF / r in terms of the ordinary filtered variables,
~
where the decomposition is given by F = F + F and is
the fluid density. Filtering, as defined above, is denoted
by the overbar and mass weighted averaging by the tilde.
Only velocity components and thermal variables are
mass-averaged. Fluid properties like density and pressure
are treated as usual.
All theories of turbulence are faced with the closure
problem arising from the basic nonlinearity of the
governing equations. Modeling of some statistical
quantities is essential to close the problem.
Direct filtering of the momentum equation yields,

~
~
t kl

D2 U k U 1
=
r
[
x l
x 1 12 x m x m
~
~
2 U1
D2 2 r 2 U k
D2
+(
)
12
2 x m x n x m x n
12
~ ~
2
2
m
D 2 r U 1U k
-(
+] + [ T
)
x m x n x n x m
x 1
12
~
~
2Uk
2 U1
mT (
+
)]
x 1 2 x 1x k
+(

~
1 U k r r
r x m x m x n
~
~
U
U 1
( k +
)+
x 1 x k

)2

viscosity m T is made up of terms quadratic in the

gradients of the filtered velocity.
The Smagorinsky model produces too much dissipation.
A second model is needed, in addition. Lee and
Meecham [3] propose a model based on a Taylor series
expansion. This model, using the first two terms, added
to that of Smagorinsky, yields for the needed SGS term
of (2). The last term is the Smagorinsky model.

SESSIONS

Lee and Meecham tested their deductive model by

comparing with a near Gaussian probability distribution
for the velocity field. Their results showed that important
correlation coefficients for the deductive model were
0.88 and 0.97 in the case of truncations up
to D 2 and D 4 , respectively, deemed excellent.

Overall Sound Pressure Level (OASPL) was

calculated, and normalized to a reference radius of 0.30
meter. Experimental acoustic data were acquired for
using a radial arc of ground plane microphones.
Measured data were normalized to 0.30 meter free field,
far field levels by correcting for spherical spreading,
atmospheric propagation and absorption, and ground
effects.
Comparisons of measured and simulated OASPLs
versus radiation angle are made in the Figure. OASPLs
for measured data include only frequencies which contain
significant jet mixing noise energy. Agreement is within
2 dB. It is emphasized that there are no adjustable
parameters. The simulation results are in excellent
agreement with the measured data in the aft quadrant
(90) where the engine noise is dominated by the jet
mixing source. Lesser angles, where internal engine noise
dominates, are not simulated.

SUMMARY AND CONCLUSIONS

Lighthill [4] developed the standard theory of aerosound

for free turbulence. For large x, the sound field is
r( x , t) - r =
2
T ij ( y , t )
[
](1 - M c cos q )- 5/2 d y

t2
4 p c4| x - y |3 V
using the spatial variation of the retarded time,
where t = t - (| x - y | / c ) is the retarded time. Here

xi x j

Mc is approximately 0.5Mj and

~ ~
2
T ij = rU iU j - tij + ( p - c r ) dij - sij.

TURBULENT FLOW FIELD AND

MIXING NOISE RESULTS
The computational model was incorporated into the
ANSWER software package developed by Analytical and
Computational Research, Inc [5]. The basic turbulence
model is the time-honored k-.
Numerical simulations were performed for a
subsonic free heated jet with operating parameters based
on experimental data for a WR19-4 mini-turbofan engine
during static operation. The circular exit area is 0.02
meter2, The engine was operated with exhaust Mach
number 0.46 and 0.78 based on nozzle exit conditions.
Turbulence quantities within the flow domain were
calculated for each of the 2048 time steps by subtracting
the mean field values from the time-dependent values.
Normalized rms axial velocity fluctuations on the jet
centerline compare well with cold jet data measured by
Lau, et al [6].

The computational structure is such that it can be

carried out on a PC in manageable time. Large Reynolds
number, Mach 0.46 and Mach 0.78 jet flow from a small
turbofan engine have been computed. Computed far field
jet mixing noise levels are in close agreement (within 2
dB) with those measured during engine field-testing. The
sound intensity is proportional to the sixth power of the
fluctuation velocity. A 2dB variation between the
computed sound using LES and the measured sound
means the LES of the fluctuation intensity has an error of
less than about 8%. The aerosound measurement,
because of the sixth power leverage, is extraordinarily
demanding for the turbulence model. The computational
procedure presented here shows great promise for noise
reduction without the need for massive experiments.

ACKNOWLEDGMENTS
The first author is grateful for support provided by
Northrop Grumman Corporations Fellowship Program,

REFERENCES
[1] Smagorinsky, J., 1963, Mon. Weather Rev. 91, p. 99.
[2] Schein, D.B. and Meecham, W.C. 2001, Am. Soc. of
Mechanical Engineers, New Orleans, May 29-June 1.
[3] Lee, C.P., and Meecham, W.C., 1984, "A Deductive
Model for Subgrid-Scale Reynolds Stress," UCLA report,
School of Engineering and Applied Science.
[4] Lighthill, M. J., 1962, Proc. Roy. Soc. A267, p. 147
[5] Runchal, A. K., 1994, ANSWER Users Manual,
Analytical and Computational Research Inc.
[6] Lau, J. C., Morris, P. J., and Fisher, M. J., 1979, Journal
of Fluid Mechanics, Vol. 93, pt. 1, pp. 1-27.

SESSIONS

Lifting Surface Hydroacoustics At High Reynolds-Number

D. A. Bourgoyne, C. Judge, J. M. Hamel, S. L. Ceccio, and D. R. Dowling
Department of Mechanical Engineering, University of Michigan, Ann Arbor, Michigan 48105, USA
The unsteady turbulent flow on or near lifting surfaces is often a source of hydroacoustic noise. Turbulence in these regions
may produce noise directly and hydrodynamically-forced structural motions may radiate noise as well. Interaction between a
hydrofoil's unsteady vortical wake and its own structure may produce undesired self-sustaining vibrations. This paper reports
on a series of recent experiments focused on these phenomena high Reynolds numbers. The tests were conducted in the United
States Navy's William B. Morgan Large Cavitation Channel at flow speeds up to 18.3 m/s on a two-dimensional test-sectionspanning hydrofoil (2.13 m chord, 3.05 m span, 0.171 m maximum thickness) at angles of attack between 1 and +1. The
measurements include foil surface static and dynamic pressures, foil vibration, LDV-determined average flow velocities and
turbulence quantities, and PIV flow fields in the immediate vicinity of the foil's trailing edge.

INTRODUCTION
Flow-induced noise may be generated when wall
bounded turbulence interacts with the trailing edge of a
flow-control surface or lift-generating hydrofoil. At
low mach number, two mechanisms dominate
hydrofoil noise production. (1) Broadband turbulentboundary-layer surface pressure fluctuations may
scatter from the foils trailing edge. (2) Vortical flow
oscillations may form in the foils near wake leading
to narrowband quadrapole sound sources whose nearfield pressures scatter from the foil as dipole sound. If
the near-wake vortex-shedding frequency coincides
with a vibrational resonance of the foil structure, a
self-sustaining flow-induced vibration may occur.
This condition, known as singing, typically enhances
noise radiation and may cause structural vibration
problems. This paper reports the findings from a new
high-Reynolds number experimental effort focused on
these phenomena. A thorough review of prior research
in this area is provided in [1].
The goal of these experiments is to provide
fundamental insight into the fluid mechanics of
trailing-edge noise generation in marine propulsion
systems at Reynolds numbers typical of actual ship
propellers (~108). In addition, these experiments will
provide a unique high-Reynolds-number experimental
database for testing and development of turbulence
and computational fluid dynamics (CFD) models.

EXPERIMENTS
The experiments were conducted at the US Navys
William B. Morgan Large Cavitation Channel (WBMLCC) in Memphis, TN. The WBM-LCC is a low

turbulence (0.2% nominal) water tunnel with a 6:1

contraction ratio and a 3.05 m by 3.05 m by 13 m test
section (Fig. 1).

FIGURE 1. Schematic of the WBM-LCC.

The test model was a cast two-dimensional

hydrofoil made of Ni-Al bronze with a 2.134-m chord
(c) and 0.171-m max thickness (t) which spanned the
WBM-LCC test section where it was centered
vertically and longitudinally. The foil cross section
was generic for Naval propellers having moderate
thickness and camber (f). The shape is that of a
NACA-16
(t/c=0.08,
f/c=0.032)
with
two
modifications. First, the bottom (pressure side) of the
foil is flat aft of 28% chord. Second, the foil terminates
with rounded bevel starting near 97% chord. The
second modification leads to a compact region of flow
separation in the vicinity of the trailing edge.
Experiments were conducted at angles of attack of
1, 0, and +1 (measured from the flat side of the
foil). The primary test speeds were 3.0, 6.0, 12.0 and
18.3 m/s yielding chord-based Reynolds number
values of 6-10 million, 16 to 20 million, 29 to 39
million, and 46 to 61 million, respectively, when water
temperature variations are taken into account. The
maximum foil lift load was measured at 730 kN.
Measurements of the flow on and near the foil were
made with an external two-component laser-Doppler

SESSIONS

velocimetry (LDV) system, static pressure taps,

dynamic
pressure
transducers,
foil-internal
accelerometers, and a two-component particle imaging
velocimetry (PIV) system. Additional experimental
details and results from the first phase of testing are
available in [2].

RESULTS & CONCLUSIONS

Normalized temporal power spectra of foil surface
pressure fluctuations, Spp, in the vicinity of the
hydrofoils trailing edge are shown on Figure 2 for
four suction-side transducers that span the location of
boundary layer separation. The normalizations of both
axes are taken from [1] (see Fig. 11-26). Each
spectrum has been truncated when the transducer noise
level begins to corrupt the signal. Here, the wake
thickness, yf, is approximately 2 cm; omega has units

of in rad./s, q is the free-stream dynamic pressure, and

Uinf is the free stream speed.
As expected, the spectra depend on transducer
location. The spectrum with the highest frequency
content and least low frequency energy occurs farthest
upstream under the attached part of the suction side
boundary layer. The spectrum showing the least high
frequency content and the greatest low frequency
energy occurs closest to the trailing edge, well aft of
the suction side boundary layer separation point.
Results are comparable at the other test speeds.
Overall, the surface pressure fluctuation spectra are
smooth and do not display a peak near (omega)yf/U
of unity that would indicate organized near-wake
vortex shedding. This suggests a lack of vortex
structure in this foils near wake in spite of its presence
in prior wind tunnel investigations of a geometrically
similar foil at lower Reynolds number [3]. The

FIGURE 2. Pressure fluctuation spectra from four flush-mounted transducers located at 93% (A), 94.5% (B), 95.75% (C), and
98.8% (D-1) chord at a flow speed of 3.0 m/s and 0 angle of attack.

vibration measurements support this contention.

Acceleration spectra were not remarkable and rootmean-square acceleration levels were less than 0.1 g.
Additional tests are planned in the near future with
different foil trailing edges.

ACKNOWLEDGMENTS
This research is sponsored by Code 333 of the
United States Office of Naval Research. Significant
technical assistance was provided by personnel from
the United States Naval Surface Warfare Center
Carderock Division.

REFERENCES
1. W. K. Blake, Mechanics of Flow Induced Sound and
Vibration, Vols. I and II, Orlando, FL, Academic Press,
1986.
2. D.A Bourgoyne., S.L. Ceccio, D.R. Dowling, S. Jessup,
J. Park, W. Brewer, and R. Pankajakshan. "Hydrofoil
turbulent boundary layer separation at high Reynolds
numbers," 23rd Symposium on Naval Hydrodynamics,
Val de Reuil, France, September 2000.
3. J. Gershfeld, W.K. Blake, C.W. Knisely Trailing Edge
Flows and Aerodynamic Sound, Paper no. 88-3826-CP,
AIAA Thermophysics, Plasmadynamics, and Lasers
Conference, San Antonio, Texas, June 1988.

SESSIONS

Acoustic Turbulence in a Rectangular Channel

S. Nomura, Y. Hayashi and K. Murakami
Department of Mechnical Engineering Ehime University, 3 Bunkyo-cho,Matuyama, Ehime,Japan
The effect of ultrasonic vibration in the Reynolds number (Re) range of 1500 to 6000 on fluid flow in a square channel
was investigated experimentally. By applying ultrasonic vibration to laminar flow which was produced by the agitation
or distubance of cavitation bubbles, the transition to turbulent flow from laminar flow was promoted downstream.
Turbulence intensity by ultrasonic vibration is larger at the sound pressure antinode due to the influence of a standing
wave. Since it is possible to control the fluid flow from outside due to the easy transmission of ultrasonic vibration in
liquids, this technique can be applied to fluid flow in various channels as a non-contact turbulence promoter.

1. INTRODUCTION
In generally, since ultrasonic energy is small
compared to the kinetic energy of fluid flow, it is
difficult to generate large-scale motion of the fluid
directly by the ultrasonic energy, and to control the
fluid flow itself. However, the direction and the flow
pattern of the fluid might be greatly changed by very
small perturbations, as is made clear from the
phenomenon of separation of boundary layers or the
turbulent transition from laminar boundary layer. In
this study we propose the promotion of turbulence by
ultrasonic vibration as a basic method of fluid control
method using ultrasonic energy. The turbulence
intensity by ultrasonic vibration was measured, and the
effect of acoustic turbulence on fluid flow was
investigated experimentally.

2. EXPERIMENTAL DETAILS
A schematic diagram of the experimental apparatus
is shown Fig.1. An acrylic square channel 10mm thick,
with a cross-section area of 50 mm by 50 mm, and a
length of 3620 mm is adopted as a test channel. The
ultrasonic transducer is fixed to the bottom of the
channel 2320mm from the entrance. The ultrasonic
Stainless
Steel Plate

Tranceducer
X=0

3620

50

50

2320
U

P ro

LDV
System

be

Wattmeter
Amplifier

Pump
Oscillator

Fig. Experimental apparatus

transducers are bolted Langevin PZT-type vibrators

with resonance frequencies of 25kHz, and ultrasonic
power in the range of 10W to 50W. Tap water was
used for the test liquid and the water flows by the
potential head difference of the tank installed on the
upstream side and the downstream side of this channel
respectively. The water temperature is limited to
12 2. The mean velocity and turbulence intensity
(rms of the instantaneous velocity deviation from the
mean velocity) were measured by a laser-Doppler
velocimeter. It is set to be x=0mm on the central axis
of the transducer in the channel, and the downstream
direction is taken to be positive. The streamwise
velocity in the channel central section was measured in
the range of 0mm<x<800mm.
The rms value of the fluctuation of flow velocity
(standard deviation) is written as turbulence intensity
in this paper regardless of the Reynolds number.

3
RESULTS AND DISCUSSIONS
Figure 2 shows local variations of the velocity
with time at y=38mm when Re = 1500. Here, y is the
distance from vibrating plate. A considerable
perturbation exists in this channel even in the laminar
flow region. By applying ultrasonic vibration, the
variation of velocity fluctuates greatly, however, the
mean velocity becomes small compared to that with no
ultrasound.
Figure 3 shows the velocity profiles and
turbulence intensity in the square channel above the
center of the transducer at 25.3kHz. The turbulence
component by ultrasonic vibration increases, and the
velocity profile is remarkably different near the
vibrating surface (y=6mm) and at y=30mm-50mm. A
great amount of cavitation bubbles were observed at
these positions Turbulence intensity with ultrasound is
larger at the sound pressure antinode due to the

SESSIONS

influence of a standing wave. The acoustic turbulence

intensity of the power output 40W is 6mm/s on
average, with a maximum of 11mm/s, and the mean
turbulence intensity without ultrasound is 2mm/s.
Consequently, there is a three-fold increase in
turbulence intensity on the average, and locally, a fivefold increase was obtained.
Turbulence intensity at the channel center ( y=25
mm) is plotted as a parameter of Re as shown in Fig.4.
Turbulence intensity becomes greater than that for no
vibration under both the Laminar flow region of less
than Re=1500 and the turbulent flow region of more
than Re=4000. On the other hand, turbulence intensity
with ultrasound is smaller than without ultrasound in
the range of Re=2500 to 3000.
Figure 5 shows the turbulence intensity measured
at y=6mm where Urms took the local maximum within
the range of Re=1500-2500. Urms with ultrasound
increases when Re is less than 1500, whereas it
Re=1500
Without ultrasound
With ultrasound ( P=40W )

Velocity [m/s]

0.1

0.05

decreases when it is over 2500.

In high Re regions, turbulence can be restrained
by the ultrasonic vibration near the wall. Acoustic
cavitation near the wall causes the reduction of
turbulence intensity under turbulent regions. These
results suggest the possibility of reduction of the wall
friction loss by acoustic cavitation. This seems to be an
analogous mechanism to the fact that turbulent
intensity is reduced when air bubbles exit along the
wall[1]. In the transition region anywhere in the
channel, the ultrasonic vibration has the effect of
restraining the turbulence of the channel flow.
Figure 6 shows the result of the comparison of
the effects of ultrasonic vibration at the downstream
region of x=800mm. Downstream, the velocity profile
approaches the turbulent velocity distribution where
the flow has a rapid velocity gradient near the wall.
Turbulence intensity hardly changes before and after
ultrasonic vibration, however, the transition to
turbulent flow from laminar flow is promoted in
downstream.
We also investigated 45kHz vibration. The result
was that, when the ultrasonic output power is the same,
a large disturbance is locally generated by 25kHz
vibration, and by 45kHz vibration, a small disturbance
is generated within the wide range in the channel.
25

0
0

10

20

Urms [mm/s]

t [s]

Fig.2 Local variation of the velocity with time, y=38mm

15
10

Re=1500

0.04

0.03

U
Urms
U ( P=40W )
Urms ( P=40W )
FIDAP

0.02

Fig.5

0.01

2000
4000
Reynolds number

0.06
0

10

20
30
y [ mm ]

40

50

Fig.3 Velocity profile and Urms in the channel, x=0mm

0.05

Re=1500

0.04
0.03

25

U
Urms
( P=40W )
U
Urms ( P=40W )
Laminar
Turbulent

0.02

Without ultrasound
With ultrasound ( P=40W )

20
Urms [mm/s]

6000

Variation of Urms with Re, x=0mm, y=6mm

U , Urms [ m/s ]

U , Urms [ m/s ]

0.06
0.05

Without ultrasound
With ultrasound ( P=40W )

20

30

0.01

15

0
10

Fig.6

5
0

2000
4000
Reynolds number

6000

Fig.4 Variation of Urms with Re, x=0mm, y=25mm

10

20
30
y [ mm ]

40

50

Velocity profile and Urms , x=800mm

REFERENCES
1. N.K. Madavan, S. Deutsch and C.L. Merkle, J. Fluid
Mech., 156, 237-256(1985).

SESSIONS

On the Study of Theory and Application of the

CoupledMode Parabolic-Equation Method
Based on the WKBZ Theory
Zhaohui Peng, Fenghua Li and Renhe Zhang
National Laboratory of Acoustics, Chinese Academy of Sciences, P.O.Box 2712, Beijing, 100080, China
An efficient, numerically robust algorithm for calculating sound propagation in the range-dependent waveguides, which is called
CMPE (Coupled Mode-Parabolic Equation), is introduced, on the base of the generalized phase integral (WKBZ) theory. The
CMPE is a hybrid model expressed in terms of the normal modes and mode coefficients. CMPE uses a PE approach in a radial
direction and normal modes in the depth direction. The Numerical calculation of a typical problem shows that CMPE has high
accuracy and fast speed. And numerical example of the broadband pulse propagation is also presented.

INTRODUCTION
Combined with the coupled mode theory and
parabolic equation method, a new model has been
developed by Abawi, Kuperman and Collins [1], that is
the coupled-mode parabolic-equation (CMPE) method.
CMPE solutions are expressed in terms of the normal
modes and mode coefficients, which satisfy coupled
horizontal wave equations and can be solved with PE
method. It is practical to apply the coupled-mode
parabolic-equation to large-scale problems and
possibly even global scale problems at low frequency.
On the progress of resolving sound propagation
problems with CMPE, the computation of local modes
and coupled coefficients is one of the most difficult
aspects, and takes most computer time. Therefore, an
efficient algorithm of the computation of local modes
and coupled coefficients is the key to improve the
efficiency of the coupled-mode parabolic-equation.
Based on WKBZ [2] theory, a new eigenvalue finding
algorithm is presented, which can calculate the
eigenvalues efficiently and accurately. By combining
the improved WKBZ theory and the coupled-mode
parabolic-equation theory, a new range-dependent
propagation model (CMPE) is presented.
The numerical results show that CMPE has high
accuracy and fast speed. The effect of boundary
variability on the transmission loss is studied in this
paper. And numerical example of the broadband pulse
propagation is also presented.

CMPE METHOD BASED ON WKBZ

The series solution of CMPE method based on
WKBZ can be written

p ( r, z ) = r

1
2

[k
n =1

( r )] 2 u n ( r ) n ( z; r )

(1)

where the local eigenvalues k n and normal modes

n ( z; r ) satisfy

2 n ( z ; r , ) ~ 2
+ k k n2 n ( z; r ) = 0
2
z
and

nm dz = nm

(2)
(3)

~
11
3
( )2 .
where k 2 = k 2 + 2
2
2 2

In this paper a fast and accurate algorithm [3] for

solving the characteristic equation Eq.(2) is used based
on WKBZ theory.
The u n in Eq.(1) is the mode coefficient, which
satisfies
!
!
u
"
= Ar u + iKu
(4)
r
!
where u = [u0 u1 u2 # u M ]T is the mode
coefficients vector. The diagonal matrix K is the
eigenvalues matrix. The coupling coefficients
matrix Ar is defined by
Ar ,i , j = i

dz .
r
Eq.(4) can be solved with numerical method.

(5)

EXAMPLES
An example on sound propagation in slope bottom
oceans is solved in this section. In Fig.1, the numerical
solutions of transmission loss calculated by CMPE,
COUPLE [4] and RAM [5] are compared. The results are
in good agreement with each other. And CMPE is
much faster than COUPLE and RAM.

SESSIONS

FIGURE 1 Transmission losses for an example on sound propagation in slope bottom oceans. The water depth is 200m at range
of 2 km decreasing linearly to 80 m at range of 8 km. The point source with 100 Hz is placed at 30m and the receiver depth is
30m. The water sound velocity is 1500 m/s, and the bottom velocity is 1600 m/s. the density ratio between bottom and water is
1.6 and the bottom attenuation is 0.5 dB/wavelength. The solid line is transmission loss calculated by CMPE taking 101 seconds
computational time. The dashed line is calculated by COUPLE with 4518 seconds, the dotted line is calculated by RAM with
5489 seconds.

FIGURE 2 Stacked time pulse vs. range. Note the splitting up of the signal with range in three distinct wave packets
corresponding to the first three modes of the waveguide. The lines are pulses at ranges of 5, 10, 15, 20, 25, 30 km form the source
placed at 10 m with bandwidth of 450 ~ 550 Hz, respectively. The water sound velocity is 1480 m/s, and the bottom velocity is
1587 m/s. the density ratio between bottom and water is 1.6 and the bottom attenuation is 0.3 dB/wavelength. (a) Slope bottom.
The initial water depth is 30m at a range of 2 km increasing linearly to 50 m at a range of 30 km. (b)Plane bottom. The water
depth is 40 m.

Another numerical example on broadband pulse

propagation is also presented in this section shown in
Fig.2.

ACKNOWLEDGMENTS
The work was supported by the National Natural
Science Foundation of China.( Grant No. 10074070)

CONCLUSIONS
EFERENCE
On the base of the generalized phase integral
(WKBZ) theory, a new approach of coupled-mode
parabolic-equation (CMPE) method is studied in the
range-dependent
waveguides.
Examples
of
transmission loss and pulse propagation have indicated
that CMPE is efficient and high precision for
range-dependent waveguides.

1.
2.
3.
4.
5.

Abawi, A. T., Kuperman W. A. and Collins M. D., J. Acoust. Soc.

Am., 102(1), 233-238 (1997)
Zhang, R., Liu, H., and He, Y., Chinese Jour. of Acous. (in
Chinese), 13(1), 1-12 (1994)
Peng, Z. and Li, F., Science in China (Series A) (in Chinese),
31(2), 165-172 (2001)
Evans, R. B., J. Acoust. Soc. Am. 74, 188-195 (1983)
Collins, M. D., J. Acoust. Soc. Am. 93, 1736-1742 (1993)

SESSIONS

Transfer Function of Structure-borne Noise to Underwater

Radiated Noise
J.-S. Kim, H.-S. Kim, H.-J. Kang and S.-R. Kim
Acoustics Research Group, Korea Institute of Machinery & Materials, Daejeon, Korea
A comparison between theoretical and measured transfer function, which relates structure-borne noise source level to
underwater radiated noise, of a naval ship is presented. Transfer functions are obtained by dividing underwater radiated noise
by the value of structure borne noise source strength below machinery mounts. In prediction, Statistical Energy Analysis (SEA)
of the whole ship structure is used to get vibration levels of hull plates. Then, far field radiated noise is calculated by summing
up radiated sound from all wetted hull plates below water line. In addition, underwater sound pressures at the distance of 1 m
away from the hull were measured to get experimental transfer functions. The two transfer functions are compared to show
reasonable agreements in spite of the subtle physical differences between each other.

INTRODUCTION

in which W is the acoustic power of radiated sound,

r and c are density and speed of sound in the
2

Underwater radiated noise (URN) prediction from

naval ships under construction is of great importance.
It is known that structure-borne noise is a predominant
noise source of URN at low speed in general.
One approach to estimate URN due to structure-borne
noise is based on the equivalent hull forces that
transmitted from the feet of equipment to the hull plate
through mountings and its seating. URN is the result of
the sound radiation from hull subjected to the point
forces, i.e. equivalent hull forces obtained above. The
other approach is to use the result of SEA, i.e. the
average vibration levels of SEA elements, in
combination with radiation efficiencies of hull
plates[1].

seawater. s rad , i , Ai , and < v > i are the radiation

efficiency, the area and the space-time averaged
velocity squared of plate i , respectively. Finally, N is
the number of plates. Note that equation (1) holds
under the assumption of individual plate vibrating
incoherently.
When applying SEA to noise prediction of ship, one
constructs the so called SEA model of the whole ship
structures. Structural plates such as hull, deck and
bulkheads together with compartments such as cabins
are idealized by thousands of inter-connected subsystems. Fig. 1 shows an example of SEA model as
seen from the bottom of the vessel.

We used the second approach to calculate URN. The

URN was divided by the value of structure-borne noise
source strength of equipment considered to yield the
transfer function. In addition, similar transfer function
was obtained experimentally by utilizing the data from
overside ship acoustical survey performed during sea
trials. The comparisons between the two transfer
functions are presented in this paper.

Prediction of urn
The total acoustic power of radiated sound into water
by vibrating plates is given by
N

W = r c s rad ,i Ai < v > i

i=1

(1)

FIGURE 1. A SEA Model of Ship Structures.

SEA computes energy densities of each sub-system by
solving SEA equations. The energy densities are
closely related with the space and time averaged

SESSIONS

velocity squared, appeared in (1), for plate members.

Therefore, we can use the SEA solutions directly to
obtain the acoustic power radiated into the water from
the wetted hull surface by using (1). As for radiation
efficiencies of plates in contact with water like ships
hull, one can use the formula suggested by Uchida et
al. [2].
Once the acoustic power is obtained, the far field
pressure is given as follow under the assumption of
spherical spreading from point source.

W = 2p r 2 p 2 / r c

(2)

After some calculations, the radiated pressure at 1 m

apart from the source is given as

L p = L w - 10 log D f + 54

FIGURE 2. Transfer Function : Diesel Generator.

(3)

where, L p is the URN level (dB ref 1 mPa, 1 m, 1 Hz),

Lw is the acoustic power level (dB ref 10-12 W),

and D f is the bandwidth.

Structure-borne noise transfer function

We defined the structure-borne noise transfer function
as

TF = L p - L a

( 4)

in which TF is the transfer function of structure-borne

noise, L p is the URN as described in (3), and La is

FIGURE 3. Transfer Function : F.O.T. Pump.

exact definitions, i.e. prediction for far field and
measurement done near hull plates. This phenomenon,
although not conclusive, could be utilized when
analyzing noise transmission path of machinery with
relatively small source levels or refining URN
prediction method in an effective way.

the 1/3 octave band acceleration level (dB ref 10-5

m/sec2) at the position below the equipment mount.
During overside acoustic surveys, a hydrophone was
placed 1 m apart from the hull to measure the
underwater sound pressure induced by the operation of
individual equipment. At the same time, accelerations
below the equipment mount were also recorded.
We calculated the experimental transfer functions
using these two measurement results, and compared
with those of theoretical values in Fig. 2 and 3 for a
diesel generator and a pump, respectively.

Concluding remarks
It is an interesting fact that the two transfer functions
compare reasonably well in spite of the differences in

ACKNOWLEDGMENTS
This work was partially supported by a grant from the
Critical Technology project of the Ministry of Science
and Technology, Korea.

REFERENCES
1. Hyun-Sil Kim, Jae-Seung Kim, Hyun-Ju Kang and Sang
Ryul Kim, An application of SEA to ship noise
prediction, NOVEM International Conference, Lyon
Congress Center, 31 Aug. 2 Sep., 2000.
2. S. Uchida, Y. Yamanaka, K. Ikeuchi, K. Hattori and K.
Nakamachi, Prediction of underwater noise radiated
from ships hull, Bulletine of the Society of Naval
Architectures of Japan, No. 686, 36-45, (1986).

SESSIONS

Flow Noise and Functional Models of

Wall-Turbulent-Pressure
E. B. Kudasheva, L. R. Yablonikb
a

Space Research Institute of Russian Academy of Sciences, 117997 Moscow, Russia

I.I.Polzunov Scientific and Development Association on Research and Design of Power Equipment (NPO
CKTI)., S.-Petersburg, Russia

Two models of spatial characteristic functional of near-wall-turbulent pressure fluctuations are suggested and analyzed in view to
develop the method of experimental investigation of characteristic functional of near-wall turbulent pressure field. They are a
Gaussian model for jet flow and a Poissonian model for turbulent boundary layer. The functional approach permits to reduce
experimental investigation of turbulent pressure field characteristic functional to measuring a limited number of parameters and
dependencies typical of the studied turbulent flow- against type.

EXPERIMENTAL METHOD OF
CHARACTERISTICAL FUNCTIONAL

and the characteristics functional is transformed into a

characteristics function of n-dimensional distribution of
probability:

From Haddle and Skudrzyk paper[1] the response of a

flush-mounted transducer to the pressure field in a
turbulent boundary layer is known as the flow noise.The
most comprehensive are continual statistical models of
spatial structure of wall-pressure-fluctuations (x)
described by characteristic
functional [(x)] =
<exp{i(x))p(x)dx}> and representing full statistical
description of a random pressure fluctuation field p(x,t).
The present work analyses forms of analytic
representation of characteristic functional of turbulent
pressure fluctuations. The functional approach developed
by the authors of the present work [2] makes it possible
to obtain an exhaustive description of random field p(x)
based on experimental investigation of characteristic
functional of wall-turbulent pressure field. Estimated
experimentally, characteristic functional [ (x)] is
defined on a set of functional arguments = K(x),
where is the sensitivity of a transducer, K (x) is its
pulse characteristics. Because of the sensor averaging
effects the characteristic function of the transducer signal
s() = <exp (is)>gives the characteristic functional:

Turbulent flows different in statistical nature of turbulent

fluctuations generation are described by different
functional models. In the case of a jet flow when pressure
fluctuations are caused by external turbulence and the
distance
between the sources and the point of
observation is not too small, the well-known integral
description of pressure fluctuations allows to represent
near-wall fluctuations as a sum of great number of
statistically independent components related to different
zones of the turbulent flow. Random values of the kind,
as a rule, have asymptotically normal distribution that
implies a model of a Gaussian field whose characteristic
functional has the form

s () = <exp {i K (x) p (x) dx}> = [K(x)].

G [ (x)] = exp {- (x1)(x2)R (x1, x2)dx, dx2}

(1)

s ( 1 .. n ) = x1 ...

xn

(1.. n)

(4)

THE CHOICE OF FUNCTIONAL

MODELS OF TURBULENT PRESSURE

(5)

For probes with rather small reception surfaces so that

pulse characteristics is reduced to delta-function K (x)
0 (x x0), the characteristic functional is : s ()
(2)
<exp {i 0 p(x0)}> = p ( 0)

Here R(x1, x2)is the correlation function R=<p(x1)p(x2)>.

The characteristic function of the signal of transucer:

corresponding to single-point probability distribution of

pressure. In the case the receiver is an array of n point
probes with sensitivity , pulse characteristics :

q (e) = K(x) K(x+e)dx is the impact function of the

receiver. The wave form of the characteristic function is
sG = exp [- 2/2S(k)E(k)dk] ,
(7)
where S(k) = q(e)exp(-ike)de is the wave characteristic

K (x) i 0 (x x i) i (x x i) ,

(3)

sG = G [l K(x)] = exp [- 2 / 2 q (e) R (e) de ] ,

(6)

SESSIONS

of the receiver, E(k) is the frequency-wave spectrum of

the turbulent pressure fluctuations.
Another functional model can be suggested for pressure
nearfield in the turbulent boundary layer. In contrast to
jet flows, near-wall pressure fluctuations in turbulent
boundary layer is essentially determined by spontaneous
splashes accompanied by surges of the liquid from the
wall towards external area of the flow. Assuming
statistical independence of the splashes and their uniform
probability distribution over the surface, one can roughly
describe the turbulent pressure fluctuations using the
Poissonian statistics :
P [ (x)] = exp {n[(g(x-y)(x)dx)1]}. (8)
Here n is mean number of splashes for the unity of area,
() is the characteristic function of the probability
distribution of pressure fluctuations P in the kernel of a
splash, g (r) is impact function of the splash (g (0) = 1)
defining spatial correlation links:
R (r) = n <2> g ( ) g ( + r) d .

[ (x)] = exp{- (x1) (x2) RG (x1, x2) dx, dx2}*

exp {nP [P ( gP(x-y) (x) dx) 1] }.

(14)

It has been show that the suggested functional approach

permits to reduce the task of experimental investigation
of the characteristic functional to measuring a limited set
of parameters typical of a given type of turbulent flows.
They are correlation function of turbulent pressure
fluctuations RG(x1,x2) for Gaussian field and the
dependencies nP, (); gP() for Poissonian component of
turbulent pressure field.

REFERENCES
1.Haddle G.P. and Skudrzyk E.J.,J. Acoust.Soc.Amer., 46, 130-157
(1969).

(9)

The signal characteristic function of homogeneous

receiver of the area of S0 (in this case, at the aperture
K(x) = K0 = const) in a Poissonian field is presented as
sP(l) = exp {n [ ( K0 S0 g (x y) dx) 1] dy}.

functional models studied in the work, or their

modifications. In the case Gaussian and Poissonian
components can be considered statistically independent,
the structure of characteristic functional [ (x)] is
obviously, within the given reasoning, a product of
expressions (5,8):

2. Kudashev E.B. and Yablonik L.R., Phys. Acoustics, 45, 467-471

(1999).

(10)

For small S0, when receiver can be considered pointshaped, we get from the above expression
sP1(l) = exp {n [ ( 0 g ( ) ) 1] d }.

(11)

If in homogeneous field, the area of the receiver surface

S0 dramatically exceeds the area 0 = g()d of the
impact zone of the spontaneous splash, then, in line with
(10),
sP(l) exp {n S0 [ ( 0 0 / S0) 1] }.

(12)

Since <>=0, it follows that for large area of the receiver

surface S0/0 >>1 and the corresponding distribution
approaches Gaussian distribution :
sP2(l) = exp [- 2 02 <P2> 20 /S0 ] .

(13)

CONCLUDING MODEL
One can assume that the variety of stochastic regimes of
wall-turbulent pressure fields, including those in
boundary layer with turbulence at its external limit taken
into account, may be represented by a superposition of

SESSIONS

The Measurement of Fluctuations of the Signal

Propagation Time over Long Acoustic Paths
A.Stromkova, I.Didenkulova, Ya.Karlikb, A.Kazarovaa, L.Lyubavina,
E.Pelinovskya
a

Institute of Applied Physics, Nizhny Novgorod, Russia

b
Morphispribor, St. Petersburg, Russia

Acoustic waves can propagate in the ocean over basin scales and therefore can serve as a unique tool for diagnostics of
ocean processes including internal waves, tsunami etc. The present paper describes results of the experiment on longrange sound propagation at the 4800 km acoustic path Hawaii-Kamchatka. Pseudo-random signals (M-sequence) at the
carrier frequency of 75 Hz were radiated by a source deployed near Hawaii. The signal frequency band was 37.5 Hz,
which allows one to measure travel time with accuracy of about 30 ms. In our measurement the phase-difference
algorithm was used. The accuracy of travel time measurement with this method was significantly improved. It was
observed that tides produce strong oscillations of travel time The results of experiments were found to be in good agree
with theoretical calculations based on the 5 model for tides.

1. INTRODUCTION
Acoustic waves can propagate in the ocean over
basin scales and therefore can serve as a unique tool
for diagnostics of ocean processes: tides, eddies,
currents, tsunami, internal waves, etc [1]. The main
physical reason of the influence of ocean processes at
acoustic signal is water motion, which results in the
change of the acoustic signal travel time. Therefore,
the measurement of acoustic travel time changes and
their identification are the main problems. Acoustics is
widely used for ocean study. It is evident that for high
spatial-time and frequency resolution high frequency
acoustic waves would be used. But sound attenuation
rapidly increases with frequency that makes
impossible the use of high-frequency sound (above
several hundred Hz) for monitoring over distances of
about several tens km. Only very low frequency sound
(below 100 Hz) can propagate over basin scales.
Traditional way to measure the travel time variations is
based on the use of the phase method and harmonic
acoustic signals. The phase is related with the travel
time through the sound velocity and the propagation
path. However, sound propagates in the ocean by
many paths rays, which produce many acoustic
arrivals at the receiver. Such arrivals usually overlap
each other and therefore can not be separated.
Consequently, the acoustic phase can give only
information on ray-averaged fluctuations for travel
time. Relatively recently more complex signal began
to use in the ocean acoustics so called M-sequence.
The M-sequence signal being long in time has wide

spectrum, and can be compressed in time by the use of

correlation technique. The M-sequence signal allows
one to measure travel time variations for different
paths or path-groups separately. In the present paper
we describe results of experimental study of sound
propagation over 4800 km. For detection of travel-time
variation the novel approach is used based of
difference phase method applied to the M-sequence
signals.

2. METHOD OF ACOUSTIC TRAVEL

TIME MEASUREMENT
Let a signal, which is the harmonic carrier modulated
in phase with the M-sequence is emitted by the sound
source. The receiving acoustic signal is the sum of
signals arrived by different paths. To analyse the
received signal one need to divide the signal by parts
of duration equalled to the M-sequence duration T. For
each of these parts defined as the k-th part of the
received signal the cross-correlation function with the
replica of the radiated signal is

bk ( ) =

tk + T

yk (t ) M ( t )dt ,

(1)

tk

where yk(t) is the k-th part of the received signal, M(t)

is the replica of the emitted signal, and T is the Msequence duration.

SESSIONS

One can introduce the function k (m) as follows

k (m) = bk ( ) bk + m ( )d =
T

= ei0 ( k k +m ) hk ( )hk + m ( )d

experimental data. The coefficient of correlation

between them achieves 0.6-0.7.
-5

(2)

x 10

Experiment
Calculation

0.8

where k is the travel time, hk () the pulse response

of the acoustic channel during k-sequence. The phase
of the function k (m) is k(m)=( k-k+m). For correct
measurements of the travel time it is necessary that the
phase k (m) for the time interval mT does not exceed
.
If is the travel time, an approximate expression for
the derivation of the travel time is

Derivation of travel time (s per s)

0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
0

d (k ) (n, k )
.

dt
2 f 0 nT

(3)

Thus, the phase difference method is based on

measurements of the phase k(m) that is the phase
variations between acoustic transmissions. The travel
time variations are directly related to the phase
variations.

3. EXPERIMENT AND RESULTS

The experiment on long-range sound propagation
was conducted at the 4800 km Hawaii-Kamchatka
acoustic path. Pseudo-random signals (M-sequence) at
the carrier frequency of 75 Hz were radiated by a
source deployed near Hawaii. The signal frequency
band was 37.5 Hz, which allows one to measure travel
time with accuracy of about 30 ms. The sound receiver
was installed near Kamchatka. Travel time fluctuations
were studied for the time period of about one year. It is
worthy to note that the signal to noise ratio at the
hydrophones in the experiment was below -10 dB
measured in the signal frequency band.
Measured fluctuations of the travel time were
compared to calculated fluctuations. The calculation
was made with the use of the global inverse tidal
model TXPO5.1 [2]. The acoustic path was divided
into 10 segment of about 500 km length. For each
segment tidal current was supposed to be constant,
which was calculated with the PXPO5.1 model for
middle point of the segment. We neglected multipath
acoustic propagation in calculations. The travel time
derivative was estimated by calculation of the travel
time in 6 minute interval. The experimental and
calculated data are shown in Figure 1.
As it is seen from Figure 1, even such a simple
model for acoustic travel time fluctuations due to tidal
currents demonstrates good correlation with the

10

15
20
Experiment number

25

30

35

FIGURE 1. The derivative of the travel time over

acoustic path. Experiment and calculation based on the
tide model.

4. CONCLUSION
In this paper the phase-difference algorithm based on
the use of M-sequence signal for detection of small
fluctuations of the acoustic travel time was described.
The accuracy of travel time measurement with this
method was significantly improved. Experiment on
long range sound propagation from Hawaii to
Kamchatka revealed that tides produce travel time
oscillations having amplitude of up to 0.2 s. The main
physical mechanism of influence of tides on travel
time is tidal currents. The proposed method can be
used for detection of water motion due to tsunami.

ACKNOWLEDGMENTS
This work was supported by RFBR (01-05-64426,
01-05-64162, 00-15-96741).

REFERENCES
1. Munk W. and Wunsch C., Deep Sea Research, 26A, 123161 (1979).
2. Egbert G. and Erofeeva S., Efficient inverse modelling of
barotropic ocean tides. Global Inverse Solution TPXO.5.1.
in http://www.oce.orst.edu/po/research/tide

SESSIONS

Granular-Hydrodynamic Model for the Convection in a

Vibrating Granular System
Guoqing Miao and Rongjue Wei
State Key Laboratory of Modern Acoustics and Institute of Acoustics, Nanjing University, Nanjing 210093, P. R.
China
we present a granular-hydrodynamic model that capture the essence of convection in a vibrating bed. A set of hydrodynamic
equations including the heat transfer are used to describe the motion of the system. The boundary condition at the bottom of the
container is taken as that the power flow equals the power dissipated by granular system during the collision complete
inelastically with the container. The solution of steady state with no flow in gravitational field is obtained. With this steadysolution (or a static distribution of temperature), a numerical simulation of a complete set of free convection equations was
performed, the result is in agreement with experiments.

The vertically vibrated granular materials exhibit

fluidlike behavior and show a variety of phenomena,
including convection, heap formation, size segregation
and surface wave [1]. Among these phenomena,
convection has attracted particular interest, because not
only can convection happen alone, but also convection
happens simultaniously together with (or within) other
phenomena (e. g. heap or wave), and might even be the
source for the wave and heap formation. The
theoretical approaches for explaining the convection
have mainly been the continuum hydrodynamic theory
[2] and large scale molecular dynamic simulation [3].
In this paper we model vibrated granular system as
''thermal'' granular fluid. Each grain takes part in
simultaneously macroscopic flow motion and
microscopic random or ''thermal'' motion, and they
have simultaneously the mean macroscopic flow
velocity and thermal velocity, which corresponds to
certain ''temperature''. Because granular system is a
strongly dissipative system, the motion of the system is
governed by momentum equation, heat transfer
equation, and continuity equation similar to that in ref.
[4].
We denote macroscopic flow velocity by u , and
mean thermal velocity by v . The momentum equation
is analogous to the Navier Stokes equation, it reads
u
p
m
+ (u )u = + u g ( v 2 )
2
t

(1)

where p is pressure fluctuation, the density of

layer, and g the gravitational acceleration. can be
called dynamic viscosity coefficient, and the
thermal-expansion coefficient, The term mv 2 / 2
represents the temperature of the system. The heat
transfer equation is

m
m
m 2

( v ) + u ( v 2 ) = ( v 2 ) I
2
2
t 2

(2)

where is the thermometric conductivity, and I

the rate of energy dissipation due to the inelasticity of
grain-grain collision. The coefficients , , and
I play a more important role here than they do in
hydrodynamics, they depend upon v [4], then
according to ref.[5], they depend upon driving
acceleration of the plate. Then Eq.(1) and Eq.(2) are
coupled not only through the term mv 2 / 2 , but also
through four coefficients. We take the density is
approximately constant, then the continuity equation is
u = 0

(3)

It is well known that in a normal fluid in a

gravitational field free convection could happen if the
externally applied vertical temperature gradient is
directed downwards and its magnitude exceeds a
certain value. In a granular system we consider that
this temperature gradient is applied by the vibrated
plate. So the problem is under what temperature
gradient the convection will happen. Then we will
solve first the problem of steady state with no
macroscopic flow (or convection) in gravitational field,
and try to find a critical temperature distribution under
which the convection will happen. For the steady state
the energy equation (2) is reduced to [4]
d 2 v 1 dv
+
v =0
dz 2 z dz
where

(4)

is a dimensionless variable given by

SESSIONS

z = ( h ' x ) / as in ref.[4]. The boundary condition at

free surface ( z = h ) is taken as
K

d 2
v
dx

=0

(5)

1/ 2

(9)

x =h

this means that the energy flux vanish at free surface.

As to the boundary condition at bottom of the container
( x = 0 ), we consider as follows. In ref.[5] we use a
model of a single sphere colliding completely inelastic
with a massive sinusoidally oscillating plate to describe
the motion of a layer of vibrated granular material, and
obtained the power input P by the plate for each grain
as a function of acceleration of the plate (shown
graphically in figure 1). So here we use

Principle, we could use the general theory for free

convection with the temperature distribution (8) to
obtain the condition under which the convection could
happen. But in fact it can hardly be done so. We can
only use numerical method to calculate it. The results
are shown that nothing happens for < 1 . For > 1
and reaches some critical value the convection does
happen. Figure 2 shows a convection pattern of two
dimensional system, which are in quality agreement
with experiment.

12

10

10

P (10-7J/s)

dI

A = p /(mKI 0 0 )
dx

6
4

0
0

FIGURE.1: (a) The mean power input, P , for each grain,

as a function of dimensionless acceleration of the plate.

10
X

15

20

FIGURE.2: A convection pattern of two dimensional

system.
this P as power input for each grain from the
boundary x = 0 , and obtain the boundary condition
K

d 1
( mv 2 )
=P
dx 2
x =0

(6)

The solution to (4) are the modified Bessel function of

zero order, I 0 ( z ) and K 0 ( z ) .
v = AI 0 ( z ) + BK 0 ( z )

(7)

Here x h corresponding to z 0 . As the function

K 0 ( z ) is singular here, v (z ) must be given by I 0 ( z )
alone, i. e.
v = AI 0 ( z ) .

(8)

This work is supported by the Special Funds for

Major State Basic Research Projects and National
Natural Science Foundation of China through Grants
No. 10074032, 19834040 and 19874029.

REFERENCES
1. H. M. Jaeger, S. R. Nagel and R. P. Behringer, Phys. Today,
49(4), 32-38 (1996).
2. Marc BourZutschky and Jonathan miller, Phys. Rev. Lett.
74, 2216-2219 (1995); Hisao Hayakawa, Su Yue, and
Daniel C. Hong, ibid 75, 2328-2331 (1995).
3. Rosa Ramirez, Dino Risso, and Patricio Cordero, Phys.
Rev. Lett. 85, 1230-1233 (2000).
4. P. K. Haff, J. Fluid Mech. 134, 401-430 (1983).
5. Guoqing Miao, Lei Sui, and Rongjue Wei, Phys. Rev. E 63,
031304 (2001).

Insert this into the equation (6), we have

SESSIONS

The Use of an Equivalent Medium for a Coupled Inversion

in Underwater Acoustics
J.C.Le Gaca, M.Aschb
a

Centre Militaire dOcanographie, EPSHOM, B.P.30316, 29603 Brest Cedex, France

Laboratoire ANAM/MNC, Institut des Sciences de lIngnieur de Toulon et du Var, B.P.56, 83162 La Valette du
Var Cedex, France

Abstract : The paper describes a shallow water geoacoustic inversion scheme based on a model based matched impulse response.
The concept is derived from a previous paper [2] which presents an equivalent medium approach. The technique exploits the most
stable part of the impulse response of the acoustic channel, giving robust estimates of geometrical and geoacoustic parameters.

INTRODUCTION
Bottom properties are essential in the frame of
shallow water acoustics, especially for very low
frequencies. Several approaches have been developed
in the last few years leading to good estimates of
geoacoustics properties. A new trend lies in the use of
broadband signals received on sparse arrays, possibly
reduced to a single hydrophone[1]. Bottom properties
have shown to be robustly and efficiently retrieved
from measured impulse responses.
This paper presents an approach following previous
work explained in [2]. The aim of this previous work
was to determine an equivalent medium and it was
applied to the INTIMATE96 sea cruise data set [3].
The idea of equivalent medium means that the
bottom thus determined behaves as the true medium in
the limit of an acoustical application (a kind of
through the sensor approach). The equivalent
medium concept consists in assessing the main
parameters when using a sonar system rather than
finding relevant physical parameters. The frame of this
geoacoustic inversion method, mainly based on an
analytical development of the reflection coefficient,
was limited to simple equivalent media (semi infinite
fluid half spaces) and questionable in the case of high
frequency dispersive true media. We propose in this
article a generalization of this previous approach.

GEOACOUSTIC INVERSION SCHEME

The INTIMATE96 data set consists of broadband
signals (300 to 800 Hz chirps, 2s duration, repeated
every 8s) received on a 4-hydrophone vertical array.
We exploit a range-independent phase shown on Figure
1. Matched filtering (cross correlation of the received
and the emitted signals) was applied. The envelopes of
the estimated impulse response thus obtained were then
extracted (Figure2). The received sequences were
divided into two parts. The first spikes are direct paths
refracted in the thermocline with a small number of
bottom reflections. They were highly sensitive to the

sound speed profile. The second part, quite stable,

exhibited a multipath structure of bottom-surface
reflected rays supposed to carry most of the
information on the bottom properties.

Inversion of Geometrical Parameters

The exact geometry of an experiment (distance
source/receivers [D], source [Zs] and receiver [Zr]
depths, water column depth [H]) is often unknown
precisely due to operating modes and the lack of direct
measurements of these parameters. For example, the
localization of the source is quite difficult when it is
towed by a boat. Geometrical parameters must be well
known in order to perform a reliable geoacoustic
inversion. Thus, we applied a generalization of a
linearized inversion scheme presented in [4] in order to
refine the a priori knowledge of the geometrical
parameters. This scheme, not explained in detail in our
paper, exploits the travel times of 12 identified bottomsurface reflected rays from the 115m depth
hydrophone. It takes advantage of the fact that these
rays are slightly deviated by the sound speed profile.
csurf=1520 m/s
zr1~35 m
H~135 m

zs~93 m

zr2~105 m
zr3~115 m

D~5.5 km
cbot=1508 m/s
cp = 1650m/s = 1.8 kg/dm
p = 0.36 dB/m/kHz

cp = 1750m/s = 2. kg/dm3
p = 0.28 dB/m/kHz

h1 = 2m

silty sand
shell,gravel and sand

cp = 3000m/s cs = 1600 m/s = 2.4 kg/dm3

p = 0.07
s = 0.25 dB/m/kHz

h2 = 0.5m

limestone

FIGURE 1. Experimental set up and environmental

parameters for the range independent leg

Results of the inversion from 20 starting models are

summarized in Table 1. Travel times obtained with
mean parameters are plotted on Figure 2 (black bars).
They match the measured travel times very well.

SESSIONS

Table 1. Inversion of geometrical parameters

Parameter
Mean (m)
Standard Deviation (m)
D
5629.5
0.4
H
134.82
0.13
Zs
90.95
0.14
Zr
115.91
0.20

Geoacoustic Inversion
Once geometrical parameters have been precisely
determined, the geoacoustic inversion can be
performed. The geoacoustic model we looked for was
a simple fluid half-space in order to compare it to the
results given in [2].

the true medium. Furthermore, the inversion was quite

stable in comparison to the one performed in [2] (for
). Except for very few localized frequencies, the
comparison of simulated coherent losses at 5.63 km
with the true medium and the equivalent one proved to
be in very good accordance between 200 and 800 Hz
(not shown in this paper). Between 50 and 200 Hz, the
same comparison showed greater differences due to the
interaction of the basement (Figure 4). The accordance
for the phase was not as good which is not surprising:
the half space fluid medium is unable to render the
same complexity of phase shift as the true medium at
bottom reflections.
1850

0.9

0.8

1800

0.7

Impulse response envelope

Cp (m/s)

0.6

Alpha (dB/m/kHz)

1750

0.7

1700

0.6

0.5

1650
0.4

0.5
1600

0.3

0.4
1550

0.3

20
40
60
80
Number of function evaluation

0.2

100

20
40
60
80
Number of function evaluation

100

FIGURE 3. Geoacoustic inversion from 20 starting models

0.2

0.1

True medium

Equivalent medium
50

50

100

100

150

150

200

200

55

0
3.75

3.8

3.85

3.9

3.95
Time (s)

4.05

4.1

4.15

FIGURE 2. Typical INTIMATE96 envelope of the impulse

response on the 115 m depth hydrophone

The algorithm is based on the comparison of the peak

levels of each bottom-surface reflected path between
the measured impulse response and a simulated
impulse response. The stable part of the impulse
response envelopes (after 3.75 s) is normalized to unit
energy, giving the following objective function to be
minimized :
f (c p , , ) = plm (i ) pls (i ) , i = 1L n

Losses (dB)

Frequency (Hz)

60

65

250

300

250

20

40

60
80
Depth (m)

100

120

300

20

40

60 80 100 120
Depth (m)

70

FIGURE 4. Coherent losses for true and equivalent media

where plm is the level of a measured peak, pls is the

level of the corresponding simulated peak, n is the
number of paths taken into account (for t>3.75s).
The simulated impulse responses were computed
with the GAMARAY ray model [5] which provides
good estimates of the impulse response in comparison
with normal mode models, even near the critical angle
(which is essential for our inversion scheme), for this
kind of environment. A parametric study showed that
the objective function had only a single global
minimum, which allowed us to use the Nelder and
Mead simplex optimization algorithm in order to find
cp and [6].
The mean estimates of the most sensitive geoacoustic
parameters were : 1678.8m/s for cp, 0.35 dB/m/kHz for
, and 1.86 kg/dm3 for . The density was deduced
from cp. Twenty starting values were chosen
stochastically
within
the
following
bounds
{[1550 ;1850]m/s and [0.3 ;0.7]dB/m/kHz} (Figure 3).
These mean estimates were in good accordance with a
weighted sum of the superficial layers parameters of

CONCLUSION
A simple and robust approach for geoacoustic inversion
has been presented in this paper. It is shown to be more
stable than previous work in [2]. It shows that the
equivalent medium is relevant in the case of
INTIMATE96 context for coherent losses between 200
and 800 Hz.

REFERENCES
1.
2.

3.

4.
5.

6.

Hermand, J.P., IEEE J.Oceanic Eng. 24,41-66 (1999)

Demoulin, X., et al., Estimating equivalent bottom
geoacoustical parameters from broadband inversion in Proc.
ECUA2000,
M.E.Zakharia
et
al.
eds.
,
Lyon,France,2000,pp.191-196
Stephan, Y., et al., Acoustical effects of internal tides on
shallow water propagation : an overview of the INTIMATE96
experiment, in Exp. Ac. Inv. Methods for Exploration of the
Shallow Water Environment, Caiti et al. eds, Kluwer
Academics Publishers (2000), pp.19-38
Dosso, S., et al, J.Acoust.Soc.Am. 104,846-859 (1998)
Westwood, E.K., et al., J.Acoust.Soc.Am. 81,1752-1761 (1987)
Press,W.H., et al., Numerical Recipes in C, Cambridge
University Press (1992)

SESSIONS

Proposal of FDTD-PE Method with Slow Varying Envelope

Approximation for Underwater Acoustic Pulse Propagation
Tetsuo Anada, Takenobu Tsuchiya, Nobuyuki. Endoh (Kanagawa University)
Toshio Tsuchiya and Toshiaki Nakamura (JAMSTEC)
3-27 Rokkakubashi Yokohama, 221-8686 JAPAN

E-mail: anada@cc.kanagawa-u.ac.jp

In this paper, a new finite-difference time-domain PE method based on the slow-varying envelope approximation is
proposed for the analysis of forward wave propagation, diffraction and reflection and given for short pulse propagation problems in underwater acoustics. The advantage of the FDTD-PEM proposed here is its simple numerical
implementation and can be analyzed within a reasonable cpu-time and memory. In order to improve the accuracy and
efficiency of the method, the computational spatial discretization of second-order differential is replaced by the
Douglas operator scheme, which the truncation error of O(x)4 is ensured in the depth direction.

1. Introduction
Parabolic Equation Methods (called Beam Propagation
Method in branches of optics), in which an acoustic field
solution can be determined by solving the one-way operator equation for the forward-propagating field, are powerful design tools for underwater wave propagation problems. A great number of PEMs have been proposed by
pioneers since its inception [1,4,5]. The advantage of the
FD-PEM is its simple numerical implementation and can
be simulated within a reasonable cpu-time and memory.
On the other hand, the method has several drawbacks because it is approximation to the Helmholtzs wave equation neglected backward-wave. In order to overcome
above drawbacks, a new approach for developing the PE
method in time-domain is to use the wave equation based
on slow-varying envelope approximation and is called
the TD-Beam Propagation Method in optics field [2,3].
Based on this SVEA, an acoustic short pulse propagation has been applied to the wave equation simultaneously
leading to an algorithm suitable for studying forward wave
propagation, wide angle of diffraction and reflection. In
this paper, we propose a very efficient time-domain parabolic equation method (TD-PEM). The aim of this paper
are: (1) to derive a new formalism for the pulse propagation in time domain; and (2) to demonstrate its possibility to treat underwater acoustic pulse propagations.

2. Basic formalism
For the sake of simplicity of formulation of TD-PEM
based on SVEA , we restrict ourselves to the two dimensional problem and the wave equation in Cartesian coordinates. For underwater acoustic wave propagation problems, the wave equation is given by
n
2

c t
2

=0

(1)

where t is time, n(r,t)=c0/c(r,t) is the refractive index, c0

and c is the reference sound speed, the sound speed in
underwater acoustics. In comparison with a carrier frequency, the signal frequency will be regarded as a slow
wave propagation. By applying SVEA to the time term
of the wave equation and substituting a solution of the
form ( r ,t )=( r ,t )e jt into Eq.(1), the following timedomain wave equation is obtained:
n
2

n
2

2 j

n
2

=+

(2)
2
2
2
2
c t
c t
c
where is the angular frequency (carrier frequency) .
The first term on the time is much smaller than the second term. With above first-order approximation, the reduced wave equation may be written as:
n n
2 j 2
= 2+ 2+ 2
(3)
c t x z
c
While, for acoustic wave propagation confined in the
x-z plane with z-propagating beam, the paraxial wave
equation is rewritten as follows:
2

2 2 2
= 2 + 2 + k 0 n n b
(4)
z x y
where nb is the reference index, k0 is the wave number.
Comparing with Eq.(3) and (4) , Eq.(3) and Eq.(4)
have the same form. It is quite obvious that PE codes
can be extended to solve the time-domain PE equation
in the slow-wave approximation. Therefore, we can be
solved by using Alternating Directing Implicit Method
(so called operator splitting method) for obtaining the
solution of Eq.(3).
The ADI method can be applied to lead to a program
suitable for Eq.(3). The principle is to use two different
equations that are used in turn over successive time-steps
each of duration t/2. That is, the first equation is solved
2

2 jk 0 n b

SESSIONS

implicitly only in the x-direction and the second only in

the z-direction. In addition, to improve the accuracy of
the simulation, the Douglas operator scheme firstly used
by Lee, et al is applied. As a result, we can easily obtain
the high accuracy six-point scheme (then, the truncation
error is the fourth-order) and lead to a tridiagonal system
of complex linear equations[6].
1
x
1
( x )
= 2
2
x ( x ) x x 1 + x

2( x, z )
( x, z) + ( x x, z )

, 0 = + +

3. Results of simulations
In order to demonstrate the validity and usefulness of
SVEA based TD-PEM, we studied the underwater acoustic pulse propagation in the upsloping seafloor model.
The depth of the channel is 100 m and the sound speed
and density in the underwater are c=1500 m/s, =1.0 g/
cm3, those of sediment layer are 1700 m/s,=1.2 g/cm3 ,
the frequency is 500Hz. An acoustic wave packet with a
depth profile corresponding to the gaussian beam is
launched at t=0 and z= 50 m. The longitudinal profile of
the wave pocket along +z propagating direction is also
Gaussian beam with full-width of 10 m. The computational window is 200 m by 1000 m which is discretized
in a 200 by 2000 grid. The time step used is 0.1E-3 [sec].
The total duration simulated is 1 [s]. The multimode channel waveguide is supported some guided modes, and is
reflected in the boundary of top and upsloping-bottom,
and the interference will be occurred. Figure 1 shows the
pulse propagation characteristics of the incident-pulse
with Gaussian profile. It is observed that the pulse is confined in the underwater column, while the part of the pulse
wave is radiated into the upsloping sediment layer. The
validity of the present method is confirmed by the coincidence with our physical image.

4. Conclusion
In this paper a recently developed FDTD-PE method
based on the SVEA has been given for predicting the pulse
propagations in underewater acoustics. In stead of CrankNicolson scheme, the Douglas operator scheme has been
implemented for the program in order to reduce a truncation error of the finite-difference approximation of the

Oct. 1996.

0
Depth[m]

2x = i 1 0 i + + i +1 ,

100
200

Release boundary condition

at top and bottom.

1000

1000

1000

0
Depth[m]

(Douglas case) , 0 (Crank - Nicolson)

12

(1)M.A Leontovich and V.A. Fock, Zh. Eksp. Teor. Fiz. 16, pp.557573, 1946.
(2) P.L. Liu, Q.Zhao, and F.S. Choa,Slow-wave finite-difference beam
propagation method, IEEE Photon., Technol. Lett. vol.7, pp.890-892,
1995
(3) G.H. Jin, et al, An improved time-domain beam propagation
method for integrated optics components, IEEE Photon. Technol. Lett.
vol.9, pp.348-350,1997.
(4) M.D.Collins, A higher-order parabolic equation for wave propagation in an ocean overing an elastic bottom, J.Acoust.Soc.Am.
86,pp.1459-1464, Oct.1989.
(5) D. Lee, A. D. Pierce, Parabolic Equation Development in Recent
Decade, J. of Computational Acoustics, Vol.3, No.2, pp.95-173, 1995.
(6)J. Yamauchi, et al, Improved finite-difference beam propagation
method basd on the generalized Douglas scheme and its application to
semivectoral analysis,, J. Lightwave Tecnol., no.10, pp.2401-2406,

100
200
0

Depth[m]

Reference

100
200

Depth[m]

second order derivative in depth direction. In addition,

by combining the higher-order Pade series expansion
with the Douglas operator scheme, it is possible that the
present method satisfies the accuracy required. In the
near future, the algorithms proposed here will be applied
in practical range dependent 3D problems.

0
100
200
1000
Range[m]
Fig.1 Pulse propagation in shallow water
with upsloping bottom.
0

SESSIONS

An Evaluation of Information Transmission Using

Underwater Auditory Sensitivity in Actual Sea Area
S. Kuwaharaa, K. Oimatsua, K. Kuramotoa, S. Yamaguchib and H. Matsuia
a

Japan Coast Guard Academy, Kure-city, 737-8512 Japan

Faculty of Engineering, Yamaguchi University, Ube-city, 755-8611 Japan

Usual divers searching and rescuing in a sunken ship have no communication apparatus because of complexity of the hull
construction. In such cases, the most simple and effective way for transmitting information is to emit acoustic signals directly
by using an underwater loudspeaker. At this time, it is fundamentally important to research in advance the following
measurements: masking of objective signals by ambient noise in the actual sea, evaluating of clearness of voice signal and tone
signal coded by Morse sign. In our research, the above measurements are examined in the actual sea area, using scuba divers
with normal hearing. Measurements of masking effects of auditory ambient noise provide data from which the excitation
pattern of the masking stimulus is derived. Evaluations of clearness of voice signal and coded-tone sign are measured by
changing output power levels of the underwater loudspeaker and distances between the loudspeaker and divers.

INTRODUCTION
Many people now enjoy marine sports such as skin
diving and scuba diving in sallow water area. Acoustic
signals through an underwater loudspeaker can be used
as a simple and effective way of preventing diving
accident. At this time, it is fundamentally important
to research in advance the following measurements:
masking of objective signals by ambient noise in the
actual sea, evaluating of clearness of voice signal and
tone signal coded by Morse sign.
In our research, the above measurements are
examined in the actual sea area, using scuba divers
with normal hearing. Measurements of masking effects
of auditory ambient noise provide data from which the
excitation pattern of the masking stimulus is derived.
Evaluations of clearness of voice signal and codedtone sign are measured by five phases of intelligibility
evaluation referring the Radio Regulation (RR) of the
International Telecommunication Union (ITU).

EXPERIMENT
The research was carried out on actual sea area nearby
Japan Coast Guard Academy, which is usual recreation
area of sea.
Measurement of Minimum Audible Field Underwater
The Minimum Audible Field (MAF) underwater was
determined by means of the method of limits.
Sinusoidal stimuli at 500, 1000, 2000 and 4000 Hz
were gated ON and OFF with a period of 500 msec.
The stimulus-generating equipments and response

units are depicted in Figure 1 consisted on equipment

common to audibility threshold experiment. The
research was carried out on two subjects who were
experienced in taking hearing test in air and in water.
Each diver was lowered to an ear depth of 2 meters.
The divers, wearing open circuit scuba equipment and
a wet suit, descended and fixed himself from the
transducer by means of a chest bar and a lead-weighted
belt. Twenty runs were made on each of the two
subjects, and an average of the ascending and
descending runs was made for subjects.
Evaluation of Acoustical Signal
As using voice signals is effective for transmitting
information to divers, acoustic signals of male and
female voice sound are emitted directly by using an
underwater loudspeaker. Divers were lowered to an ear
depth of 2 meters, and distances between divers head
and the projector separated were of 20, 50, 100, 160
and 200 m. Output levels of the projector were
adjusted adequately by using attenuators
The clearness of voices at each distance were
evaluated by six phases which are referred to the
following intelligibility based on RR of ITU: 5
excellent, 4 good, 3 fair, 2 poor, 1 bad and 0 no signal.
The subjects consisted of two divers who had
experienced as a radio operator on Japan Coast Guard
vessel.
It had been predicted that voice signals were masked
easily by ambient noise of experimental sea area.
Therefore, tone signals coded as QF of the Morse
sign were adopted because coded tones were not
comparatively masked by noises.

SESSIONS

Analog
Gate

Power
Amp

Attenuator

Underwater
Loudspeaker

DAT
Sound
Level
Meter
2m

5
l=20m
l=50m
l=100m
fitting

4
Evaluation of Cle

Signal
Generator

Mic

0
80

100

120

dB

140

160

140

160

(a) Female Voice Signal.

5

FIGURE 1. Setup for underwater hearing-acuity measurement.

Minimum Audible Field Underwater

In Table 1, some experimental results estimated for
minimum audible field (MAF) underwater are reported
(dB re 1 mPa). The MAF levels in quietness measured
by the previous research [1] are expressed in this table.

Evaluation of Cle

RESULTS

l=20m
l=50m
l=100m
fitting

0
80

100

120

dB

(b) Tone Signal

As low frequency ambient noises are large (about 95

dB/Hz) in the experimental sea area, measured MAF
levels rise compared with quietness levels [2,3]. On the
other hand, in high frequency region, MAF levels do not
rise very much. Furthermore, it is predicted that MAF
levels increase clearly in high frequency region when a
diver was wearing a diving hood [4].
Evaluation of Clearness of Acoustic Signals
In Figure 2 is shown the evaluation of clearness
(intelligibility) of the female voice signal and tone
signal varying signal levels at divers head points
received. The clearness of voice signal is rapidly
dropped with the signal level (slope of dashed line is
steep). On the other hand, the clearness of tone signal
is slowly dropped (slope is gentle).
In Figure 3, the evaluation of clearness of
tone signal varying distances is presented. Data from
four frequency sequences of increasing distance with
fixed source power are shown.

FIGURE 2. Evaluation of Clearness (Intelligibility) of

(a) Female Voice Signal, and (b) Tone Signal.
5
freq (source power)
500Hz (145dB)
1 kHz (150dB)
2 kHz (150dB)
4 kHz (150dB)

4
Evaluation of Cle

Table 1. Levels of minimum audible field (dB re 1 mPa).

Frequency
No Hood With Hood
In Quietness
500 Hz
99 dB
100 dB
78 dB
1 kHz
100 dB
99 dB
86 dB
2 kHz
88 dB
100 dB
89 dB
4 kHz
94 dB
119 dB
91 dB

0
10

Distance (m)

100

1000

FIGURE 3. Clearness of Tone Signal with Fixed Source

Power Levels.

REFERENCES
1. K.
Oimatsu,
K.
Kuramoto,
S.
Kuwahara
and
S. Yamaguchi, J. Marine Acoust. Soc. Jpn, 21, 103-109 (1994)
(in Japanese).
2. J. E. Hawkins and S. S. Stevens, J. Acoust. Soc. Am., 22, 6-13
(1950).
3. S. S. Stevens and H. Davis, Auditory Masking, Fatigue, and
Persistence, in Hearing, New York, Acoust. Soc. Am., 1938,
pp.208-224.
4. S.
Kuwahara,
K.
Oimatsu,
K.
Kuramoto
and
S. Yamaguchi, J. Marine Acoust. Soc. Jpn, 25, 86-91 (1998) (in
Japanese).

SESSIONS

Estimated Model of the Dynamic behaviour of a

Cavitating Valve
V. Villouvier
EDF/DRD/AMV 1 Avenue du Gnral de Gaulle 92141 Clamart France
The Residual Heat Removal circuit for the French PWR sections comprises two valves which regulate the hot and cold flow
rates. These valves operate according to a cavitation system when the circuit is functioning under certain conditions, and the
considerable excitations which are then created may lead to the cracking of certain small lines located nearby.
In order to limit the circuit's vibrations, optimal operating conditions are being sought which would make it possible to reduce
the pressure fluctuations at source-level.
The valve vibration model which has been developed links the internal acoustic energy to the vibrations measured over the valve
body using local hydraulic variables. Measurements taken on-site in respect of a great many control configurations have made it
possible to readjust the different modelling parameters.
Within the context of the RRA circuit, the local valve vibrations are estimated according to the variables governing the global
operation of the circuit, by using the hydraulics calculation and taking into account the different hydro-acoustic sources
simultaneously present.

CONTEXT
Control description and parameters

The hydraulic balance in that case involves the

aperture angle of 13VP (see Figure 1).

On-site tests
The vibratory measurements were carried out with the
aim of analysing the behaviour of the circuit and the
control components in regard to a large number of
operating conditions and, in particular, the sensitive
cases of vacuum conditions (low upstream
pressures).

THE VIBRATION MODEL

The sources of excitation

12VP

13VP

01DI

26VP

FIGURE 1. Diagram of part of the RRA circuit

The RRA circuit of the French PWR units includes

two hot and cold flow rate control valves. Under
certain operating conditions, these valves, together
with the bypass diaphragm, are in cavitation operating
mode and the considerable excitations which are then
generated may lead to the cracking of certain small
lines situated nearby.
We are seeking optimal operating conditions enabling
the pressure fluctuations at the level of the sources of
excitation to be reduced in order to limit the circuit
vibrations.
The circuit's main operating parameters are: the total
flow rate, the upstream pressure, opening of the 12VP
valve and the open or closed state of the 26VP valve.

Hydro-acoustic identification tests carried out in the

laboratory made it possible to study the equivalent
internal sources responsible for the vibrations
observed on the valve casings.

FIGURE 2. Example of acoustic sources for a cavitating

valve

SESSIONS

The Thoma number enables the degree of cavitation

Pav - Psat
to be estimated: s =
Pam - Pav

The vibrations of the 13VP valve result from

superimposing the different excitations present:
3
2
d = a A ' i Ph i
i =1

Description of the model

On the whole, it suffices to readjust five coefficients

which are the constants for the vibration model
(Figure 3).

Adimensionalisation
Excitation is adimensionalised by the head loss, the
cross-section of flow of the fluid in the valve, the
flow rate in the valve and the ratio of upstream and
downstream pressures.
We finally obtain the expression:
If s> s lim
f>1 Hz Dq= A Qadim Stb
If s< s lim
If 1<f<fc Dq= A Qadim Stb
If f>fc Dq= A Qadim Stb(fc)
With
Qadim= F(DP, x D/D0, Q, Pam/Pav)

Vacuum operation
13 VP displacements
2000

1500
Displacements

Two physical phenomena were highlighted

The turbulence noise, especially low frequency
energetic, is represented by a spectrum in the form
of:
Dqturb= A Stb
with
A: proportionality coefficient
3
fD px(1 - x )
St: Strouhal number: St =
4Q
b: constant
x= h/D
The cavitation noise is shown by the appearance
of a plateau beyond an fc frequency which varies
with the intensity of the phenomenon.
The weaker s, the weaker fc and the greater the
amplitude of the plateau.

Model
13VP measurement
1000

500

123bf10910 bo0910 vo01b

45678910 11 12 vo008b vo006b vo004b vo002b vo10700 vo10800 vo1082 vf10700 bar1

13 14 b31210 b3120

15 b25

N essai

FIGURE 3. Example of calculated, measured vibrations

Estimated calculations
Estimated calculations made it possible to highlight
the effect of the circuit's operating parameters on the
vibratory levels of the valves and helped in the search
for new control rules that would stress the circuit as
little as possible (figure 4). These rules are at present
being successfully applied by the operators.
Vacuum

700 m3/h

13VP displacements
800
26VP closed

700

26VP open
Displ measured O

600

Integration

f2

f1

Dq 2 df

The vibration model is based on the hypothesis that

the internal acoustic power is proportional to the root
mean square value of the displacements measured
directly at right angles with the excited valve.

APPLICATIONS
Hydraulic and vibratory comparisons
Initially, the hydraulic conditions corresponding to
the test configurations were calculated and compared
with the measurements taken at different points of the
circuit. In fact, the sources of excitation depend on
the local hydraulic conditions which must be
accurately estimated.

500
Displacements

The hydro-acoustic power Ph emitted is obtained by

frequency integration of the Dq curves, i.e.:

400
300
200
100
0
0

10

15

20

25

30

Angle 12VP

FIGURE 4. Vibrations calculated on 13VP, according to

the opening of 12VP, in regard to two circuit configurations

REFERENCES
1. A. BOYER - J.F. LAURO Etude hydroacoustique dune vanne
papillon en rgime cavitant - Colloque dhydrotechnique
156me Session du comit Scientifique et Technique de la SHF
- Chatou, 19 et 20 /11/97
2. J.P. TULLIS Hydraulics of pipelines Pumps, valves, cavitation,
transients Wiley Interscience

SESSIONS

Underwater Acoustic System with a Small Antenna for the

Detection and Bearing of Sound Sources
R. Salamon, J. Marszal, M. Rudnicki
Technical University of Gdansk, Department of Acoustics, Narutowicza 11/12, 80-952 Gdansk, Poland
Today's passive sonars usually detect and determine the direction of sound sources using linearly placed hydrophones. In
practice, however, there is a need for passive systems whose antenna is as small as possible. To meet these requirements the
array will usually consist of four ultrasonic transducers located in the corners of a square and the additional fifth one placed in
the centre of the square. The subtractions of the signals from the opposite transducers are proportional to the sine and cosine of
the angle of the arrival wave. The accuracy with which the angle of the coming wave is determined is low because taking away
the signals reduces the useful signal more than does noise. The output signal to noise ratio is worse than the input one, which
has a negative impact on the accuracy of the bearing. Because the signal to noise ratio gets worse as the system's operating
frequency band gets wider, the narrow band filtration as a way of improving the accuracy of the bearing is used.

SYSTEM OPERATING PRINCIPLE

For long M series and the above conditions, the

spectra determined have the following form:

The system's antenna is built of four hydrophones

located in the vertexes of a square with a 2 d diagonal
[1]. The signals received by the hydrophones can be
written as:
s 1 ( t ) = u( t + c ) + n 1 ( t )
s 2 ( t ) = u( t + s ) + n 2 ( t )

(1)

s 3 ( t ) = u( t c ) + n 3 ( t )
s 4 ( t ) = u( t s ) + n 4 ( t )

(3)

S ( k ) 4U ( k ) + N ( k )
where k=0,1,2,...,M-1 is the number of the spectral
line, k =cM/kfs.
The last steps of signal processing consist of the
following operations:
Y ( k ) = Im[ S c ( k ) S ( k )]

where c = ( d / c ) cos , s = ( d / c ) sin

and n1(t),...,n4(t) are the hydrophone output noises and
c is the acoustic wave velocity.
Following subtraction in pairs and summation,
three signals are obtained:
sc ( t ) = s1 ( t ) s3 ( t )
ss ( t ) = s2 ( t ) s4 ( t )

Sc ( k ) 4 j ( d / k )U ( k ) cos + N c ( k )
S s ( k ) 4 j ( d / k )U ( k ) sin + N s ( k )

(2)

s( t ) = s 1 ( t ) + s 2 ( t ) + s 3 ( t ) + s 4 ( t )
The signals are filtered in low-pass filters which
limit their spectrum to the fM, frequency. Next, they are
sampled at frequency fs to be processed in an ADC
converter into numerical series sc(m), ss(m) and s(m) at
M length. These series are used to determine discrete
Fourier transforms in the computer, further denoted as
{ } . The system assumes the following inequalities:
fs c<<1 and fs s<<1. The inequalities mean that
distance 2d between the hydrophones is much smaller
than the smallest wavelength M =c/fM in the spectrum
of the signal being received.

X ( k ) = Im[ S s ( k ) S ( k )]

(4)

The points that determine the direction of wave

arrival concentrate around the point determined with
no noise present. The scattering of the points and
consequently the measuring error are more or less
attributed to the noise in the system. The particular
noise spectrum lines are represented by points located
around the centre of the co-ordinate system. (Fig. 1).
For constant amplitude of the sinusoidal signal, the
area denoted by these points gets smaller as the signal
to noise ratio grows. Also, as the signal to noise ratio
goes up the spread of the points determining the
direction of the arrival wave gets smaller. When the
S/N ratio is greater than 10 dB the area of the noises is
practically the same as that of the co-ordinate system
centre. Through the system it is possible to determine
the direction of sinusoidal arrival waves of various
frequencies. For the same signal amplitudes, the points
that correspond to lower frequencies are located nearer
the co-ordinate system centre. The system also works
correctly for periodical signals of varying periods.

SESSIONS

ERRORS IN MEASURING THE WAVE

INCIDENCE ANGLE
To estimate the measuring error of the arrival wave
angle let us substitute relations (3) to formulas (4).
Following transformations, we obtain:
Y ( k ) = 4 p k | U ( k ) | cos + 4 Im[ N c ( k )U ( k )] +

+ p k Re[ U ( k )N ( k )] cos + Im[ N ( k )N c ( k )]

X ( k ) = 4 p k | U ( k ) | sin + 4 Im[ N s ( k )U ( k )] +
+ p k Re[ U ( k )N ( k )] sin + Im[ N ( k )N s ( k )]

L2 ( k i )

2 ( k i )
L

(10)

Variance 2(ki) can be determined using formula

(8). If we assume that the random variables in this
formula are not correlated, we get:
2 ( ki ) N 'y2 ( ki ) + N 'x2 ( ki ) = 2 N 'x2 ( ki ) = 2 k2i

value and variance 2(ki), we can assume that the

variance of the error's mean value computed from
formula (9) is approximately equal to:

(5)

where pk=4 d / k.
The first components of the sum in both equations
describe the components of the useful signal, while the
others describe the interference components responsible for the wrong measurement of the angle . With no
noise present (Nc=Ns=N=0) for any signal the distance
between the points is Zs(k)=4pk|U(k)|2, and the angle
( k)= . By denoting the useful components as Uy(k)
and Ux(k), and noise components totals as Ny(k) and
Nx(k) formulas (5) can be reduced to:

(11)

The standard deviation of the bearing error can be

finally given in the following form:
L ( ki )

2 2
pki U i LM

(12)

where - standard deviation of noise.

The standard deviation of the bearing error
diminishes as the S/N ratio, the number of spectral
lines and averaging increase. The formula derived
above was proved both in detailed studies and
measurements taken in the field.

Y ( k ) = U y ( k ) + N y ( k ) , X ( k ) = U x ( k ) + N x ( k ) (6)
The bearing error [rad] can be written as:
tg[ ( k )] =

N 'y ( k ) sin N x' ( k ) cos

1 + N x' ( k ) sin + N 'y ( k ) cos

where
N x' ( k ) = N x ( k ) / Z s ( k ) ,

(7)

N 'y ( k ) = N y ( k ) / Z s ( k )

Values 1 / N x' 2 ( k ) and 1 / N 'y2 ( k ) describe the

signal to noise ratio for the particular spectral lines.
The system in question works correctly only for
a relatively high S/N ratio, i.e. for ki spectrum lines
those correspond to the basic frequencies of the useful
signal spectrum: N x' 2 ( k i ) << 1 and N 'y2 ( k i ) << 1 . For
those spectrum components formula (7) is reduced to
this form:

( ki ) N ( ki ) sin N ( ki ) cos [rad] (8)

The mean bearing angle error is:
1 L
(9)
( ki ) = ( ki , l )
L l =1
Because of insufficient data, the error denoted with
formula (9) is merely a weak estimate of the mean
value in a statistical sense.
The bearing error ( ki) can be estimated using
formula (8) and by including the averaging operation
given in formula (9). If we assume that the bearing
error is a Gaussian random variable with a zero mean
'
y

FIGURE 1. Example of a bearing signals representation.

'
x

REFERENCES
1. R. Salamon, J. Marszal, A. Raganowicz, M. Rudnicki.,
Application of Fourier Transformation in a Passive
Sonar with Gradient Hydrophones., Proc. of 5th ECUA,
Lyon 2000, Vol. 2, pp. 1115-1120.
2. T. K. Moon, W. C. Stirling, Mathematical Methods and
Algorithms for Signal Processing, Prentice-Hall, Inc.,
New Jersey 2000.

SESSIONS

Krill vertical migrations in the Ross Sea

Kalinowski J., Azzali M., Godlewska* M., Lanciani G.,
Sea Fisheries Institute, Ancona, Italy
* International Centre for Ecology PAS, Dziekanw Leny, Poland
Krill vertical migrations in the Ross Sea were studied on the basis of three expeditions during the Antarctic summers of 1989/90,
1994/95 and 1997. Although there was no light change in the area during this time of a year krill did change its depth within a
24 hour cycle. The depth of krill swarms was averaged over two hour time intervals and approximated by the function H(t)=Ho +
A1 cos (2t/T1+1) +A2 cos (2t/T2+2), where Ho is the mean depth of krill, A1 and A2 are the amplitudes of migrations with
periods of T1=24 hour and T2=12 hour accordingly, 1 and 2 are the phases of the migration, determining the time when
migrations start. The migration pattern differed between years, with no migration in 1989/90, inverse migration in 1994 and
classical migration in 1997. The amplitude of migrations was increasing with the length of individuals and with the latitude. The
24 hours periodicity was best pronounced for large adult krill, while for juvenes 12 hours cycle dominated.

Depths of krill swarms were determined from the

echo charts as the lower plus the upper limit of a swarm
divided by two. These values were averaged over two
hour time intervals and approximated by the function:

INTRODUCTION
The phenomenon of zooplankton diel vertical
migrations is one of the most common and best
documented types of animal behaviour. The hypothesis
that changes in animals vertical distribution are caused
by the rhythmical changes in incident solar radiation
had been put forward already in the last century [1] and
documented many times since than.
Data on migrations of Antarctic krill are
contradictive. Some authors have observed migrations
[2, 3, 4, 5], others did not [6, 7].
In the present work krill vertical distributions were
studied in situation when there is no change of light
during the 24 hour period, as observed in the Ross Sea
during the Antarctic summer. The aim was to check if
krill are migrating in such conditions, and if they do,
what are the factors influencing the migration pattern.

H(t)=Ho+A1cos(2t/T1+1)+A2cos(2t/T2+2)
where Ho is the mean depth of krill occurrence, A1 and
A2 are the amplitudes of migrations with periods of
T1=24 hour and T2=12 hour,1 and 2 are the phases
of the migration, determining the time when migrations
start.
The 24 and 12 hours periodicities were assumed as
typical for krill migrations [5, 8]. Parameters of this
model Ho, A1, A2, 1, and 2 were estimated with
nonlinear regression using Marquards [9] iterative
algorithm to minimize the residual sum of squares.

RESULTS AND DISCUSSION

MATERIALS AND METHODS

During all three expeditions to the Ross Sea krill

were distributed mainly in the upper 70 m, with very
few swarms below 100 m (Fig.1).
Animals were performing diel vertical

Data on krill swarms vertical distribution were

collected during three Italian Expeditions to the Ross
Sea, in 1989/90, 1994/95 and 1997.

1994

1 9 90
100

200

100

num ber of sw arm s

200

10

10

30

30

30

50
70
90
110
130

50
70
90
110
130

depth interval

10

depth interval

depth interval

1997

n u m b e r o f s w arm s

n u m b e r o f s w ar m s

1 00

2 00

50
70
90
110
130

Figure 1. Krill vertical distributions during three Italian expeditions to the Ross Sea

SESSIONS

December 1997

12

time (hour)
0

16

20

24

24

20

24

Large krill
time (hour)
0

mean depth
experiment

40

20

40

model depth

30

16

20

20

12

16

50

depth (m)

depth (m)

12

60

time (hours)
0

Small krill

depth (m)

migrations with different pattern each year. In 1997 it

was so called normal migration, i.e. krill were
closer to the sea surface during the night and deeper in
the water column during the day, in 1994 it was
reversed, krill were closer to the surface during the
day and deeper at night, while in 1989/90 season no
clear difference in krill distribution between day and
night was detected (Fig. 2).

60

20
40

December 1994

60

time (hour)

depth (m)

12

16

20

24

30

Fig. 3. Migration pattern for small and large krill.

40

The different cycle of migration for large and small

krill is in agreement with data received by Godlewska
[5] for the Western Antarctic. She also found that the
mean depth of krill occurrence and the amplitude of
migrations depend on food availibility. Unfortunately
only in 1994 data on phytoplankton were collected
[11]. The lowest concentrations of chlorophyll were
observed at most northern stations, and they increased
southernly. This increase of food availability with the
latitude was accompanied by increase in the intensity
of krill migrations.

50

model depth
experiment
mean depth

60
70

January 1990
time (hour)
0

12

16

20

24

depth (m)

20
30
40
50

model depth
mean depth
experiment

60

Figure 2. Krill migration pattern in different years

Only during the 1994 and 1997 expeditions

representative net samplings were received for the area
of acoustic survey. In 1994 the sizes of krill were
similar in the whole area [10] while in 1997 a distinct
separation of adult and young animals in different areas
was observed, thus enabling comparison of migration
patterns for strongly different size of animals. There
was a cluster of juveniles with the lengths of 20-29
mm, and a cluster of adults, where the length class of
40-49 mm predominated. Comparison between
migration patterns for these two areas (Fig. 3) shows
that small juveniles migrated with 12 hour periodicity,
while large adults within a 24 hour cycle. The mean
depth of krill occurrence and the amplitude of
migration did not differ significantly.

REFERENCES
[1] Cuvier G. La regne animale 2 (Poissons),
Paris, 532 pp. (1817)
[2] Kalinowski J. Pol. Arch. Hydrobiol., 25: 573-583
(1978)
[3] Tomo P. Ber. Polarforsch. 4: 191-195 (1983)
[4] Godlewska M, Klusek Z. Polar Biol. 8: 17-22
(1987)
[5] Godlewska M. Pol. Arch. Hydrobiol. 43: 9-63 (1996)
[6] Daly K.L., Macaulay M.C. Mar. Ecol. Prog. Ser. 79:
37-66 (1991)
[7] Loeb V.J.,Shulenberger E. Polar Biol. 7: 363-373
(1987)
[8] Pavlov V.Ya. Tr VNIRO 66:104-116 (1974)

[9] Marquardt D. J. Soc. Ind. Appl. Math. 11: 131-136

(1963)
[10] Azzali M., Kalinowski J., In: Ross Sea Ecology
Faranda F., Guglielmo L., Ianora A. (eds).
Springer-Verlag, Berlin Heidelberg 433-455 (2000)
[11] Saggiomo V., Carrada G.C., Mangoni O., Ribera
dAlcala M., Russo A., J. Mar. Sys. 17:115-127 (1998)

SESSIONS

Comparison of Decision Tree and Multistage

Neuro-Fuzzy Classifiers of Seafloor using Wavelet
Coefficients of Acoustic Echoes
T. V. Dung, M. Moszynski and A. Stepnowski
Department of Remote Monitoring Systems, Technical University of Gdansk
ul. Narutowicza 11/12, 80-952 Gdansk, Poland
In recent years the neuro-fuzzy expert systems have been successfully applied to seabed classification from acoustic
echoes [1]. In particular, implementation of multistage Incremental Fuzzy Neural Network architectures (IFNN)
demonstrate good performance and high classification rate, especially when the number of input parameters was
reduced by Principal Component Analysis (i.e. to the wavelet coefficients extracted from sea bottom echoes only).
The paper presents the comparison of the IFNN system with the other approach, utilising most recently developed
decision tree classifier, which constructs classification models by revealing and analysing patterns found in seabed
echo records.

INTRODUCTION
In last decade the advanced swath-beam techniques
using multibeam sonars have been successfully
introduced for imaging and classifying the seabed.
However, the conventional methods of normal
incidence utilising bottom backscatter from a singlebeam echosounder are still in use, due to their
simplicity and versatility. Among these methods the
application of expert systems and neural networks
have been justified its practical usefulness [1].
The
paper
investigates
the
neuro-fuzzy
classification system in comparison with decision tree
classifier, using the same input data i.e. wavelet
coefficients extracted from sea bottom echoes.

INCREMENTAL FUZZY NEURAL

NETWORK CLASSIFIER
The first classifier analysed in this paper is based on
ANFIS artificial neural network fuzzy inference
system model [3]. ANFIS is able to derive the optimal
shapes of membership functions and number of fuzzy
rules from the given data sets. The structure of fuzzy
inference subsystem is hidden in the neural
network, therefore the system adapts its parameters in
the learning process.
The ANFIS was implemented in the multistage
incremental architecture IFNN, as seen in Fig. 1.The
input variables have been divided into M sets and
each of them is fed to an individual reasoning stage.
Therefore there are totally M single-stage ANFIS
models in serial structure and the fuzzy inference is
carried out stage by stage [3]. Variable y(k) (k<M) is
the intermediate variable which represents the output

from stage k as well as the input to stage k+1. The

decision is later fine-tuned by considering more and more
factors until the final decision, corresponding to the
output variable, are undertaken [3].

FIGURE 1. Basic structure of the Sugeno IFNN adopted

for a multistage system

DECISION TREE CLASSIFIER

The decision tree algorithm generates a classifier in the
form of a decision tree structure. This is either a leaf,
indicating a class or a decision node that specifies some
test to be carried out on a single attribute value, with one
branch and subtree for each possible outcome of the
test. A decision tree can be used to classify a case by
starting at the root of the tree and moving through it until
a leaf is encountered. At each nonleaf decision node, the
case's outcome for the test at the node is determined and
attention shifts to the root of the subtree corresponding
to this outcome. When this process finally reaches a leaf,
the class of the case is predicted to be that recorded at
the leaf.
If any algorithm can be said to have fundamental
importance in this software, it is the process of
generating an initial decision tree from a set of training
cases. However, the tree-building process is not
intended merely to find any such partition, but to build a

SESSIONS

tree that reveals the structure of the domain and thus

has predictive power. Ideally, we would like to choose
a test at each stage so that the final tree is simplest.
One of the methods for constructing a decision tree
from a set T of training cases is gain criterion, which
is defined as [2]:
gain(X) = info(T) info X(T)
where info(T) represents in fact the entropy of the set
T, info X(T) is defined as i(|Ti|/|T|) info(Ti) and |T| is
number of cases in set T.
The gain criterion prefers the attributes, which have
higher gains. Although the gain criterion gave quite
good results, it has a serious deficiency, as it
generates a strong bias in favour of tests with many
outcomes. By analogy with the definition of info(T),
we have split info = -i(|Ti|/|T|) log2(|Ti|/|T|). This
represents the potential information generated by
dividing T into n subsets, whereas the information
gain measures the information relevant to
classification that arises from the same division. Then,
gain ratio(X) = gain(X) / split info(X)
expresses the proportion of information generated by
the split that is useful, i.e., that appears helpful for
classification. The gain ratio criterion selects a test to
maximise the ratio above, subject to the constraint
that the information gain must be large at least as
great as the average gain over all tests examined.

records and its generalisation ability was checked on

testing data set of 1361 records. Sample decision tree
diagram is shown in Fig. 3. As it was shown in the Fig.
4, the decision tree algorithm works satisfactorily and
allows achieving good classification results: 92.22% .

FIGURE 3. Decision tree obtained from C4.5 program.

For comparison, concurrently to decision tree

algorithm (where all input parameters were processed
simultaneously) the incremental fuzzy neural network
was also investigated [1]. In the IFNN architecture, all
input parameters were processed sequentially and their
sequence was determined by the PCA. The final
percentage of correctly classified echoes obtained was
93.02% as seen in Fig. 5. Results achieved from both
investigated methods are similar. However, decision tree
computation time is shorter in comparison with IFNN.
1 0 0 %

RESULTS

80%
60%

In both investigated classification methods the

Wavelet Transform have been used, as the Principal
Component Analysis (PCA) showed, that its
coefficients demonstrate higher degree of importance
in seabed classification performance than the other
echo parameters [1]. The wavelet coefficients Ci are
defined by Discrete Wavelet Transform (DWT):
Ci = C( j, k ) = n x( n)j, k ( n)
where j,k=2 (2 n-k) wavelet filter constructed
from wavelet function ( ). Experimental data was
acquired from acoustic surveys carried out in the
Southern Baltic using a single-beam digital
echosounder DT4000 of operating frequency 200
kHz. Four types of sediments were represented in the
collected data: type1 - mud, type2 - fine and mediumgrained sand, type3 - medium-grained sand and
type4 - gravel, hetero-grained sand and rock. A set
of the following parameters was extracted from each
digitised bottom echo: eight first wavelet coefficients
Ci and sums of the absolute values of wavelet
coefficients of jth level S j . The program C4.5 [2] was
used to build up the decision trees. Decision tree
was trained on a learning set of 182 bottom echo
-j/2

-j

40%
20%
0 %

Type1

Type2

Type3

Type4

FIGURE 4. Testing results of the decision tree algorithm;

echoes correctly classified: 92.22%.
100%
80%
60%
40%
20%
0%

Type1

Type2

Type3

Type4

FIGURE 5. Testing results of the IFNN system; echoes

correctly classified: 93.02%

REFERENCES
1. Moszynski M., Dung T.V., Stepnowski A.: "Analysis
of the Influence of Wavelet Coefficients ... ", Proc. of
the Fifth, ECUA'2000, Lyon, 2000, vol. 1, pp.301-306.
2. Ross Quinlan J., "Program for machine learning",
Morgan Kaufmann Pub., London, 1988.
3. Dung T.V., Stepnowski A., ACUSTICA - acta
acustica, 86, 830-837 (2000).

SESSIONS

Impulse Radiation of a Sound at Condensation of a Vapor

Cavity in a Liquid
N.A.Pribaturin, M.V.Alekseev
Institute of Termophysics SB RAS , Novosibirsk, 630090, Russia
The experiments of pressure impulse radiation under collapse of vapor cavity in liquid under drastic direct contact between cold
water and vapor are suggested.

INTRODUCTION
 

There are a plenty of hydroacoustic problems where

one of the sources of nonlinear acoustic perturbation is
a vapor cavitation. It is well known [1] that a collaps
of vapor cavity produces in liquid a pressure pulse.
The asymptotic case of such behavior of a vapor cavity
is the solution of Rayleigh problem [2]. For these
problems, the initial perturbation for cavity collapse is
the difference between pressure in the vapor and the
liquid. It was proven experimentally [3] that even a
slight pressure disturbance in the equilibrium system
liquid plug - vapor plug can generate a high-amplitude
pressure pulse in liquid. Moreover, this process may
have a resonance character [4].
The objective of this work was an experimental
study of possibilities for generation of a powerful
pressure pulse after drastic direct contact between
cold water and vapor (with equal initial pressures).
The factor of disturbance in this case must be the
temperature drop.

EXPERIMENTAL SETUP
Let us consider a volumes of a cold liquid and hot
vapor divided heat-insulated membrane (Figure 1). The
pressure in both volumes is identical, and temperature
of a liquid is much less than vapor temperature. In turn
the temperature of vapor corresponds to temperature
of saturation at the given pressure. Lets imagine, that
the membrane sharply disappears so, that a cold liquid
and vapor instantly enter into contact with each other.
According to thermodynamic equilibrium condition, the
vapor pressure begin to decrease to the saturation level,
which corresponds to the initial temperature of the
liquid. Our tests demonstrated that this is true : just
after contact a rarefaction wave penetrates into vapor. It
has two zones: the zone of slow pressure decrease and
the zone of a drastic pressure fall. During that, the vapor
temperature falls below the saturation temperature and
small droplets emerge in the vapor. Due to the produced
pressure drop, liquid follows to the vapor side. The
liquid velocity is the higher, the higher the temperature
drop between liquid and vapor.

  

  

"$# % & ' (

N

)$* * ( & ' (
  


  !

FIGURE 1. Scheme of pressure impulse radiation and wave

diagram.
+

If the liquid has some additional load- P , (e.g.,

hydrostatic load), then the liquid velocity may be
even higher. After disappearance of vapor (complete
condensation on drops and fragments of cold liquid
penetrating into the vapor zone), the initially accelerated
liquid column (Figure 1) slows down abruptly on a solid
wall. This, according to the momentum conservation
law, produces a pressure wave.
To perform that kind of experiments, a special setup
was designed and assembled according to a scheme of
a shock wave tube. This is a vertical tube which divide
on two part by means thin diaphragm. Its scheme are
shown on Figure 1. The bottom and top parts are filled
the vapor and cold liquid accordingly. For maintenance
of constant initial temperatures of vapor and liquids
was used liquid thermostats. A loading of a free level
of a liquid made the break of membrane. Experiments
were performed for water, with temperature in the
range from 35 to 96 - , the vapor temperature was
103 - . The initial pressure in vapor and liquid was 0.01
MPa. During experiments, the pressure and temperature
for both liquid and vapor, and liquid velocity were
measured.

RESULTS OF EXPERIMENT
A typical pressure profile measured directly on the
solid wall is shown in Figure 2. We see that after
membrane rupture it takes some time for liquid to be

SESSIONS

accelerated and vapor to condense. After that, at the

moment of liquid deceleration, a powerful shock wave
emerge. It penetrates with the sonic velocity upward
the liquid column. After reflection from the free surface
(Figure 1), it transforms into a rarefaction wave. Thus,
a powerful pressure pulse emerges: its shape is depicted
in Figure 3.
The maximal duration of the pulse is determined
by the double time of run for the pressure wave along
the liquid column. At that, we observe conservation
of momentum - the further it is from the deceleration
point, the less the pulse width and higher its amplitude.
The amplitude of pressure pulse depends on the length
of the liquid plug- l , length of the vapor cavity - l ,
, temperature of liquid -T and pressure on the free
surface.
A considerable role in dynamics of the cavity
belongs to mass transfer between vapor and liquid.
By simultaneous measurement of the pressure and
the volume of vapor cavity, we measured the heat
flux on the liquid-vapor interface. This value is by
thousand times higher than for the equilibrium state.
The possible reasons for this effect are a strong
development of interface, cooling of vapor fragments,
strong turbulization of liquid and cooling of vapor
cavity due to formation of droplets of cold liquid.
Behavior of vapor being compressed by moving
liquid cannot be described by adiabatic law. Moreover,
the vapor pressure falls below the initial level. The
liquid velocity and amplitude of pressure pulses can
reach 20 - 40 m/s and 20 - 40 MPa correspondingly and
the specific
energy of a pulse can be about 500 - 900


kJ/m . The dependency of the maximal pressure pulse
amplitude (at the wall) on a dimensionless complex is
plotted in Figure 4. The complex includes the lengths of
the liquid and vapor plugs, and the total pressure drop
for the liquid plug. Here we present also the amplitude
calculated from a simple equation of liquid motion with

R 6 $CT

233



7,7+8

011
.//
* ((
(
#$ ####

%& %%%'

() ((+* (

() ((,* 45

FIGURE 3. Typical profile of pressure impulse.

!
!
Water, P  =0.1 MPa, T =60 C, T " =103 C, l =0.45
l  =0.37 m
t uv w

x y

m,

N OPN Q R STN UVWQ NYX Z[ V X

\]Y^ _ ` a b cT` d ^ ` e^ d b a _ f
g hij k l m n

MJK
MKK
LJK
LKK
JK
I

9 :;

< =>

? @A

B CD

E FG

E FH

o z y { | p q r s
} ~ 

FIGURE 4. Dependence of pressure amplitude from

parameters of experiments

using of momentum conservation law. The calculation

is a good description of experimental points.

ACKNOWLEDGMENTS
This work was supported by Russian Foundation for
Basic Research (grant No. 00-02-18004-)

REFERENCES
[1] I. Aya and H. Nariai, Occurrence threshold of pressure
oscillations induced by steam condensation in pool
woter, Bul. of JSME, Volume 29, No. 253 (1986)

R   $CT


[2] R. T. Knapp, J. W. Daily and F. G.

Hammitt, CAVITATION, McGRAW-HILL BOOK
COMPANY(1970)

 

 

[3] V.E. Nakoryakov, B.G. Pokusaev, N.A. Pribaturin,

S.I. Lezhnin, Behaviour of a vapor-liquid medium in
non-stationary dynamic conditions, Heat Transfer 1994:
Proc. 10th Int. Heat Transfer Conf., Brighton,(1994),
Volum 3, 389-393.

GLDSKUDJP



GHVWUDFWLRQ


 

  

 

 

  


FIGURE 2. Pressure profile on solid wall.

!
!
Water, P  =0.1 MPa, T =60 C, T " =103 C, l =1.1 m,
l  =0.37 m

[4] N.A. Pribaturin, M.V. Alekseev, V.A. Fedorov,

Resonanse phenomena in complex vapour condensation
in cooling tube, Letter J. Techn. Physics,(2000), Volum
26, No.14, 13-16.

SESSIONS