See discussions, stats, and author profiles for this publication at: [Link]

net/publication/3232163

Underwater Acoustic Communications Using Time Reversal

Article in IEEE Journal of Oceanic Engineering · November 2005

DOI: 10.1109/JOE.2005.862137 · Source: IEEE Xplore

CITATIONS READS
179 1,345

6 authors, including:

Geoffrey F Edelmann Sungju Kim

United States Naval Research Laboratory HYUNDAI MOBIS
66 PUBLICATIONS 1,201 CITATIONS 14 PUBLICATIONS 412 CITATIONS

SEE PROFILE SEE PROFILE

Tuncay Akal
University of California, San Diego
135 PUBLICATIONS 3,379 CITATIONS

SEE PROFILE

All content following this page was uploaded by Tuncay Akal on 14 November 2012.

The user has requested enhancement of the downloaded file.

852 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 4, OCTOBER 2005

Underwater Acoustic Communications

Using Time Reversal
G. F. Edelmann, H. C. Song, Member, IEEE, S. Kim, W. S. Hodgkiss, Member, IEEE, W. A. Kuperman, and T. Akal

Abstract—This paper contains theoretical and experimental I. INTRODUCTION

results on the application of the time-reversal process to acoustic
communications in order to improve data telemetry in the ocean.
A coherent underwater acoustic communication system must deal
with the inter-symbol interference caused by the time-varying,
T HE field of underwater acoustic data telemetry is rela-
tively new and has changed substantially since the intial
deployments of incoherent communication systems. Today,
dispersive, shallow-water ocean environment. An approach is these systems transmit critical though simple commands but
demonstrated that takes advantage of the focal properties of time
are too slow for large data transfers such as those required for
reversal. The spatial and temporal compression available at the
time-reversal focus mitigates channel fading, reduces the disper- video transmissions and networks. Coherent communication
sion caused by the channel, and increases the signal strength. systems more efficiently use the limited bandwidth available
Thus, a time-reversal communication system does not require in the ocean waveguide, though dispersive and time-varying
spatial diversity at the receiver, i.e., an array of receiving sensors, multipath channel characteristics prevented their practical
but takes advantage of spatial diversity at the transmitter. The
application until the early 1990s [1].
time-reversal communications system concept is demonstrated
using experimental data collected in shallow water. Data telemetry More recent work in underwater phase coherent communi-
bit rates of 500 bps (BPSK) and 1000 bps (QPSK) with bit error cations succeeded in simultaneously tracking phase as well as
rates of 0 out of 4976 bits and 254 out of 9953 bits, respectively, mitigating multipath [2], [3]. This approach coupled a decision
were obtained when transmitting to a receiver at a distance of 10 feedback equalizer with a second order phase lock loop. Newer
km, with a carrier frequency of 3500 Hz, and a 500 Hz bandwidth.
communication systems are incorporating higher frequencies
In a shallow-water upslope region, bit error rates of 15 out of 4976
bits and 14 out of 4976 bits were achieved over the same distance. and different forms of diversity. In general, higher frequency
In neither case was complex processing at the receiver used (i.e., systems facilitate higher bit rates but are more susceptible to
channel equalization, error correction coding). Time-reversal environmental fluctuations and attenuation. By using large ar-
transmissions are intercompared with single source and broadside rays of sensors, larger bandwidths, or redundantly transmitting
transmissions and shown to have superior results in both range
at different times, communication systems make a trade-off be-
independent and dependent bathymetries. The time-reversal
performance appears limited by self-generated inter-symbol in- tween complexity and stability in a time-varying environment.
terference. In addition, an initial look at the application of a single Time reversal, or phase conjugation in the frequency domain,
channel adaptive channel equalizer to received time-reversal is a process whereby a source at one location transmits sound
communication sequences is presented. The same properties that which is received at another location, time reversed, and retrans-
are beneficial to a single channel receiver are also beneficial to
mitted. The retransmitted sound then returns to its original loca-
adaptive channel equalization. A single channel RLS DFE equal-
izer is cascaded with the received time-reversal sequences and tion. Parvulescu and Clay first implemented time reversal in the
shown to further reduce scatter in the I/Q plane. The bit error ocean over 40 years ago using a single element receiver/trans-
rate decreased in all but one of the cases. mitter pair [4], [5]. More recently, substantial advances have
Index Terms—Adaptive channel equalization, communications, been made using time reversal in an ultrasonic laboratory set-
decision feedback equalizer, phase conjugation, telemetry, time ting where an array of receiver/transmitters called a time-re-
reversal. versal mirror (TRM) has been used to focus sound [6], [7]. The
application of time reversal arrays to the ocean was proposed
theoretically in the early 1990s [8]–[11]. Simulations showed
the potential use of a water column spanning array of acoustic
Manuscript received February 5, 2004; revised May 27, 2005; accepted July source/receive elements to focus energy without a priori knowl-
24, 2005. This work was supported by the Office of Naval Research under Grants edge of the ocean environment. Subsequently, the time reversal
N00014-94-1-0458 and N00014-00-1-0094. Associate Editor: J. Preisig.
G. F. Edelmann was with Marine Physical Laboratory, Scripps Institution process was demonstrated experimentally in a shallow water
of Oceanography, San Diego, CA 92093 USA. He is now with U.S. Naval ocean environment [12].
Research Laboratory, Washington, DC 20375 USA (e-mail: [Link]- The application of time reversal to communications was first
mann@[Link]).
H. C. Song, W. S. Hodgkiss, and W. A. Kuperman are with Marine Physical discussed in a laboratory setting [13]. Ocean acoustic simu-
Laboratory, Scripps Institution of Oceanography, San Diego, CA 92093 USA lations then demonstrated time-reversal communications for a
(e-mail: hcsong@[Link]; whodgkiss@[Link]; wak@[Link]). 3500 Hz pulse with a 500 Hz bandwidth [14]. Finally, the first
S. Kim was with Marine Physical Laboratory, Scripps Institution of Oceanog-
raphy, San Diego, CA 92093 USA. He is now with Agency for Defense Devel- experimental verification of time-reversal communications took
opment, Korea (e-mail: sikim@[Link]). place at sea [15], [16].
T. Akal was with NATO Undersea Research Centre, La Spezia, Italy. He Similarly, passive phase-conjugation communication sys-
is now with SUASIS, TUBITAK-Technological Development Zone, Gebze-
Kocaeli, Turkey (tuakal@[Link]). tems were proposed that implement the time-reversal process in
Digital Object Identifier 10.1109/JOE.2005.862137 computation. Instead of using a TRM actively to retransmit the
0364-9059/$20.00 © 2005 IEEE
EDELMANN et al.: UNDERWATER ACOUSTIC COMMUNICATIONS USING TIME REVERSAL 853

measured probe source signal, the receive array alone can be th element of the TRM, an array of source/receiver trans-
used to correlate the channel response and the communications ducers. Exact knowledge of the TRM transducer positions is
sequence [17], [18]. In essence, a passive phase-conjugation unnecessary since time reversal naturally corrects for an arbi-
system transmits a one-way communications sequence as trary array shape. The TRM then retransmits the time reversed
would be done in a typical single source system. However, version of the signal, . Thus time reversal implements a
before the data is transmitted, a channel probing symbol is linear matched filter yielding the autocorrelation of the impulse
broadcast to measure the multi-element channel impulse re- response of the waveguide. The signal received back at the orig-
sponse. That vector channel response is correlated with the inal source location, , can be written as:
array data and the results combined across channels. In this
literature, time reversal as discussed here is referred to as active
(1)
phase conjugation.
Time reversal is an approach that can be used independently
of or in conjunction with adaptive channel equalization. The Where the summation term has been called the Q-function in
temporal compression of the time-reversal focus reduces the [18]. The strength of the time-reversal focus depends on the
dispersion caused by the channel. The spatial focus of the number of multipaths received and the number of TRM ele-
time-reversal mirror (TRM) mitigates the effects of channel ments .
fading and provides a high signal-to-noise ratio. The receiver in The spatial resolution and sidelobe suppression issues for
a time-reversal communication system does not require spatial time reversal are the same as they are for linear broadband
diversity, i.e., an array of receiving sensors. Instead, it takes matched field processing [20] since matched field processing
advantage of spatial diversity at the transmitter. Thus, a TRM is the computational implementation of time reversal [21]. A
could be used in conjunction with a relatively simple receiver simple way to estimate spatial resolution is to recognize that
possibly including a single-channel adaptive equalizer, an a water column-spanning array can produce a vertical focus
advantage over other approaches requiring relatively complex on the order of the water depth divided by the number of
multichannel adaptive equalization. Furthermore, since the contributing modes. In many environments the highest modes
time-reversal focus degrades slowly over time as the ocean correspond to the later multipaths arriving in the impulse
changes, the inclusion of an adaptive channel equalizer could response. Receiving more multipaths indicates that the higher
extend the period of reliable communication before needing to modes are strongly contributing to the pressure field and a
capture a new probe source (PS) ping estimating the channel sharp resolution can be achieved [22]. A dense array reduces
response to each element of the TRM. the spatial sidelobes of the time-reversal focus [23]. From the
This paper will discuss results from a May–June 2000 ex- modal point of view, a denser TRM samples the water column
periment in which a source/receive array broadcast to a vertical sufficiently to capture and retransmit the higher order modes
receive array (VRA). Two-way time-reversal transmissions are that contribute to sidelobe suppression.
compared to both one-way traditional single source and broad- The TRM records and retransmits both signal and ambient
side transmissions. Data telemetry bit rates of 500 bps (BPSK) noise [26], [27]. However, retrofocused noise is incoherent and
and 1000 bps (QPSK) are presented with measured bit error disperses while the coherent time-reversal signal focuses at the
rates (BER) of 0/4976 and 254/9953, respectively. In a shallow- probe source position. Thus noise has little effect save to con-
water upslope region, BER of 15/4976 and 14/4976 were mea- sume transmit power at the TRM. Noise and attenuation also
sured. In both cases the receiver was located at a distance of limit the number of multipaths captured at the TRM.
10 km, with a carrier frequency of 3500 Hz, and a 500 Hz band-
width. Section II provides an overview of time-reversal theory B. Experimental Setup
and the experimental setup. Then Section III intercompares the The Focused Acoustic Fields 2000 (FAF-00) experiment was
results from single source, broadside and time reversal trans- conducted 19 May–15 June 2000 in a shallow water region north
missions. Time reversal is shown to have the lowest bit error of the Elba Island, Italy as shown in Fig. 1. Transmissions were
rate. Section IV will present the initial results of cascading an recorded both in a relatively flat region where water depth varied
adaptive channel equalizer with the received time-reversal com- from 110–120 m and in the slope region closer to the shore.
munication sequences. Lastly, concluding remarks are made in Fig. 1 also shows a two-layer geoacoustic model of the environ-
Section V. ment [24], [25]. The 2–3 m sediment layer is composed of sand.
The reflective bottom leads to extended multipath.
The experimental setup is shown in Fig. 2. The probe source
II. BACKGROUND THEORY AND EXPERIMENT (PS) has a center frequency of 3.5 kHz which corresponds to
a wavelength of 0.43 m in the ocean. The time-reversal
A. Theory of Time Reversal
mirror is an array of 29 source/receive transducers (SRA) with
The theory behind time reversal [8]–[11] has been presented an inter-element spacing of 2.78 m for a total aperture of 78 m
earlier. Time reversal can be seen as implementing a matched or 181 [28]. The transducers are centered at 3.5 kHz and have
filter of the impulse response of the waveguide [19]. When a a bandwidth of 1 kHz. The time-reversal transmissions were re-
known signal is transmitted from a source in a waveguide, ceived by a vertical receive array (VRA) which had 32 elements
it is convolved with the impulse response of the waveguide, with 3-m spacing. Three probe sources at depths of 40, 60, and
and the signal is received on the 80 m were collocated with the array. Data from the TRM were
854 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 4, OCTOBER 2005

Fig. 1. The FAF-00 experiment took place north of Elba Island, Italy. During the initial data collection, the TRM was moored in 110–115-m deep water. The
probe source was collocated with the VRA which was located 10-km north. Similarly, during the range-dependent portion of the experiment the VRA was closer
to shore in about 40-m deep water. Measurements of the downward refracting sound speed profiles were taken from the CTD casts from the R/V ALLIANCE.
Thermistor chains TC1/TC2 and conductivity, temperature, and current meter moorings CTCM1/CTCM2 took samples at 10 min and 30 sec intervals, respectively.
Also shown is a two-layer geoacoustic model for the area [24], [25].

radio telemetered back to the R/V ALLIANCE while the VRA 1 min after the original PS transmission. The backpropagated
data were cabled. The position of the VRA is shown in Fig. 1 signal will refocus at the original PS location where the VRA
for the range-independent portion of the experiment. can measure the vertical and temporal structure of the field. For
Collocated with the VRA, the PS transmits a short 2 ms pure the purpose of comparison, the receiver closest to the PS will be
tone pulse. The pulse is received 10 km away on the SRA, dig- used as our single element receiver. During the range-dependent
itized, time-reversed (phase conjugated) and retransmitted. The experiment the SRA was moored while the VRA was suspended
turn around time from PS transmission to TRM retransmission from the ship. During the range-independent experiment both
is over 1 minute. Thus, all time-reversal foci shown are at least the SRA and VRA were moored.
EDELMANN et al.: UNDERWATER ACOUSTIC COMMUNICATIONS USING TIME REVERSAL 855

Fig. 4. Block diagram for the generation of a time-reversal symbol.

Fig. 2. Time-reversal experimental setup. 1) A probe source (PS) transmits a 60-m and 25-m depth, respectively, for range-independent and
pulse of sound. 2) The time-reversal mirror, an array of source/receiver (SRA)
elements, records the complicated arrival structure. 3) The received signal is range-dependent propagation measured 4 June at [Link] UTC
time reversed and retransmitted into the ocean. 4) The acoustic energy converges and 9 June at [Link] UTC. Due to the reflective bottom, the
to the original PS depth and range. A vertical receiver array (VRA) is collocated 2 ms pulse was spread to over 15 ms of dispersive arrivals in the
with the PS to measure the backpropagated field structure.
range-independent region. The center column shows the broad-
side transmissions measured 1 June at [Link] UTC and 9 June
at [Link] UTC, respectively. The right column shows time-re-
versal foci measured 4 June at [Link] UTC and 9 June at
[Link] UTC, respectively from top to bottom. The channel
response measured in the left column was time reversed and re-
transmitted with about a 1-min turn around time. Note that two
dead transducers on the TRM are visible in the left column. Also
note the vertical axis scale change in the range-dependent trans-
missions measured at the VRA. The water column was about
40-m deep. In the upslope region, the time-reversal focus has a
vertical extent of less than 2 m. The range-dependent PS trans-
mission was made from the PS collocated with the VRA in
shallow water to the TRM in deeper water.

D. Time-Reversal Communication Theory

Here we will present a time-reversal communications system
Fig. 3. Comparison of the three transmission types: Single source, broadside concept. This phase coherent system transmits phase encoded
and time reversal. The first column shows how the original 2 ms pulse has information from the TRM to a point receiver. In this section, no
dispersed from single transducers at about 60-m and 25-m depth, respectively, in
both range-independent and range-dependent environments with the reception array processing or adaptive channel equalization will be con-
recorded at the SRA. The second column shows reception at the VRA of a 2 ms sidered at the reciever. The VRA element closest to the PS posi-
broadside transmission from the SRA. The last column shows the time-reversal tion is assumed as the desired target for communications. Thus,
foci measured on the VRA for both environments.
the receiver was located at 60-m depth in the range indepen-
dent case and 25-m depth in the range dependent case. First the
C. Measured Transmission Types transmission process is described and then the receiver structure
Three types of transmissions were carried out in this exper- is presented.
iment: single source, broadside and time reversal. One-way A simple phase coherent phase shift keying system was im-
acoustic communications from the TRM to the VRA either plemented. Binary phase shift keying (BPSK) and quadrature
can be carried out via a single transducer or by a broadside phase shift keying (QPSK) sequences were transmitted with bit
transmission in which all elements of the TRM transmit the rates of 500 bps and 1000 bps, respectively.
same sequence on all elements simultaneously (i.e., broadside In a time-reversal communication system, first the desired
beam). The time-reversal transmission is a two-way method target excites the channel and the channel response is measured
as shown in Fig. 2. First, a PS pulse is sent to the TRM which by the TRM. Data transmission is suspended during the PS ping
measures the channel response. The TRM then time-reverses capture period. The entire measured channel response (e.g., the
and retransmits the channel response back to the VRA. first column of Fig. 3) is used to create the communication sym-
Fig. 3 shows an example of each transmission type where the bols. In the case of BPSK, there are two symbol states that en-
underlying pulse was 2 ms in duration. The upper three panels of code one bit of information. Each element, , of the time-re-
the figure show range-independent measurements made over a versed channel response is multiplied by either to define the
10-km path in about 110–120-m water depth. The lower panels two symbol states. These two vector waveforms are stored and
show measurements made in the range-dependent upslope en- used to generate the transmitted communication sequence.
vironment over a 10-km path. The left column shows plots of The phase shift keying system illustrated in Fig. 4 encodes
the channel response as recorded on the TRM. The PS was at time-reversal symbols for an arbitrary number of bits, , on
856 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 4, OCTOBER 2005

Fig. 5. Block diagram of the single element PSK receiver system. The received signal and noise are separated into their in-phase and quadrature components.
The output of the integrate-and-dump processor (T is the symbol period) is the position of one estimated symbol in the complex plane. The output then is passed
on to a decision device which decides the most likely symbol sent.

states. To generate a set of time-reversal symbols, the III. COMMUNICATIONS RESULTS

time-reversed channel response is decomposed into its in-phase, This section will discuss measured time reversal, broadside
, and quadrature, , components and then resynthesized and single source communications results obtained in the range
with a new phase, , and amplitude, . For a symbol length independent and range dependent bathymetry north of Elba
and symbol number , the communication sequence transmitted Island.
by channel of the TRM is:
A. Communications-Range Independent Environment
First, communication sequences transmitted over 10 km in
(2) the flat-bottom region north of Elba Island are considered. The
same sequences were transmitted using the three transmission
The decomposition and resynthesis is done over all elements of types at a symbol rate of 500 symbols/sec. Recall that the re-
the SRA for all possible states. ceiver is a single element collocated with the PS.
The single element receiver structure is shown in Fig. 5. The performance results for the three transmission types are
Ocean acoustic transmissions are subject to phase distortions shown for BPSK and QPSK encoding in the second and third
that corrupt coherent communications. Before a communi- columns of Fig. 6, respectively. The first row of each transmis-
cation sequence can be demodulated successfully, the phase sion type displays the scatter plots of the estimated data sym-
distortion must be corrected. A phase locked loop (PLL) is bols. The real and imaginary outputs of the receiver for each
a nonlinear adaptive phase tracking device that produces an symbol are depicted as dots in the complex plane. Without noise
estimate of the phase distortion [29, Chap. 6.2]. The distortion or ISI there only would be two points for BPSK (180 degrees
may be caused by changes in transmitter-receiver range or apart in the complex plane) or four points for QPSK. The second
environmental fluctuations. Phase tracking is absolutely essen- row shows the calculated mean squared error (MSE) for the
tial in a coherent PSK communications system. The estimated BPSK and QPSK cases. The error is calculated by taking the
phase distortion then is removed from the received signal as difference between the actual symbol and the soft decision esti-
part of the quadrature downconversion process shown in Fig. 5. mate of the symbol. The MSE is smoothed by taking the average
This process basebands the signal thereby removing the 3500 over 200 symbols.
Hz carrier frequency. The signal is filtered, integrated over the Basic one-way communications was carried out by a single
period of one symbol, then sampled. The output of the receiver TRM transducer transmitting to a single receiver of the VRA.
(every ) is the position of a single estimated symbol The 60-m deep transducer of the SRA transmitted the 10 sec se-
in the complex plane. A symbol decision then must be made. quences to the VRA. This transmission type is the most prone to
The logical decision for BPSK is the closest Euclidean distance multipath, channel fading, as well as obtains the worst SNR and
to and for QPSK and . is shown in the first two rows of Fig. 6. The SNR is calculated
A simple carrier recovery scheme was implemented. The over the 500 Hz bandwidth. The first panel clearly shows that
squaring procedure incorporated into the PLL reduces the SNR the energy is dispersed with depth. Only the first 2 sec of the
available for PLL tracking. For the purpose of phase tracking, BPSK sequence is shown since it and the QPSK sequence qual-
QPSK data is squared twice to remove the quickly fluctuating itatively are identical. At the desired receive depth of 60 m, the
phase encoding and therefore suffers more SNR loss than observed SNR of about 6.5 dB was relatively high. A bit error
BPSK data which is squared once. In addition, inter-symbol rate (BER) of 218/4976 was obtained for the BPSK sequence
interference (ISI) affects detrimentally the ability of the PLL to and 4896/9952 for the QPSK. The PLL was unable to track
track the phase distortion [30]. phase for the QPSK transmission for the reasons discussed ear-
Communication sequences must be synchronized accurately lier. The BPSK transmissions were made on 4 June at [Link]
at the receiver or else performance can be degraded severely. UTC and the QPSK transmissions were made at [Link] UTC.
This paper assumes proper synchronization. The phase and am- Ideally, the broadside transmission will excite preferentially
plitude of the first symbol (e.g., or for BPSK) also was the first few waterborne modes in the waveguide. Exciting ex-
assumed known to avoid phase ambiguity. actly a single mode in shallow water experiments has proved
EDELMANN et al.: UNDERWATER ACOUSTIC COMMUNICATIONS USING TIME REVERSAL 857

Fig. 6. Comparison of received single source, broadside, and time-reversal range independent communication sequences. The first column depicts the first 2 sec
of measured BPSK data on the VRA located 10 km from the SRA. The second and third columns show scatter and mean squared error (MSE) plots of the
BPSK and QPSK sequences, respectively. The (in-phase and quadrature) scatter plots provides a qualitative indication of the performance of a single element
integrate-and-dump receiver located at the same depth as the PS.

quite difficult and requires more complicated feedback control modes means fewer multipaths and therefore less ISI. There is
systems [31]. For the purpose of communication, exciting fewer little dispersion and high SNR at the depth which corresponds
858 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 4, OCTOBER 2005

Fig. 7. Comparison of received single source, broadside, and time-reversal range dependent communication sequences.

to the peak (largest amplitude) of the first mode. Off peak, the tailed a priori environmental knowledge. The same communica-
other modes contribute significantly to the total field leading to tions sequence transmitted by the single source was transmitted
dispersion and ISI. As with time reversal, a broadside transmis- simultaneously on each SRA transducer with equal power as
sion uses the spatial diversity of the TRM without requiring de- shown in the third and fourth rows of Fig. 6. The energy of
EDELMANN et al.: UNDERWATER ACOUSTIC COMMUNICATIONS USING TIME REVERSAL 859

the broadside transmission was largest at the deepest element The broadside transmission was used for comparison with
of the VRA near the peak of the first mode. At the communica- the time-reversal sequence because it also used the spatial
tions depth of interest (60 m) the received SNR was 21 dB and diversity inherent in the SRA. Compared with time reversal,
the BER was 89/4976 and 829/9952 for BPSK and QPSK, re- the broadside communications sequence transmitted about 14.6
spectively. These sequences were transmitted 4 June at [Link] and 14.8 times the energy per symbol in the range-independent
UTC and [Link] UTC. and range-dependent environments, respectively. This estimate
The fifth and sixth rows of Fig. 6 show the results of the was made by noting that the broadside transmission transmitted
time-reversal communications approach which yielded the best at peak power across all 29 elements for 10 s. The multi-ele-
results. The received SNR was 19 dB and the BER was 0/4976 ment time-reversal communications sequence was normalized
and 254/9952 for BPSK and QPSK, respectively. The gray over all time and TRM transducers. Some captured noise also
level plot of measured acoustic energy shows that the focus was present in each time-reversal symbol affecting the total
is at the desired depth of 60 m and therefore helps ameliorate transmit power. Nevertheless, this is a realistic aspect of the
channel fading. The reduced scatter is due to more than just retransmission process. Although the time-reversal symbol is
improved SNR but to temporal compression (reduced ISI) as longer in time, it rarely transmits at peak power at any given
well. It should be noted that for the BPSK and QPSK time-re- time and depth. The net effect is that about 11.6 and 11.7 dB
versal transmissions, an interval of 7 and 9 min, respectively, more power is transmitted in the broadside case. However, the
elapsed between PS capture at the SRA and communications SNR received at the depth of interest is only about 2 dB higher
sequence retransmission. The channel response was captured for the range independent broadside transmission over the
on 4 June at [Link] UTC and the communication sequences time-reversal transmission (see Fig. 6). In the range-dependent
were retransmitted at [Link] UTC and [Link] UTC. environment, the broadside transmission had a SNR 3 dB lower
than the time-reversal transmission.
B. Communications-Range Dependent Environment C. Time-Reversal Gain
Communication sequences also were transmitted in the ups- Fig. 8 shows the theoretical BER for a signal measured in ad-
lope region off the northern coast of Elba island. As shown in ditive white Gaussian noise [32, Chap. 7]. Ideally, the time-re-
Fig. 3, the channel is characterized by severe and extended mul- versal process would ameliorate all of the ISI caused by the
tipath on the order of 30 ms. During this portion of the exper- channel response. The multipath in the single source results and
iment, the VRA was deployed over the side from the R/V AL- self-generated temporal sidelobes in time-reversal results cause
LIANCE which was anchored in 40-m water depth. The VRA the BER to fall above the curve. To better compare the time-re-
was at a range of 9.8 km south of the TRM. The short array had versal and the single source transmissions, white Gaussian noise
19 elements with 2-m spacing and a total aperture of 36 m. The was added to the time-reversal reception such that both commu-
PS was deployed at 25-m depth and nearly collocated with the nication sequences have equal SNR in the 500 Hz band. Neither
VRA. sequence falls on the theoretical line, however the time-reversal
The range dependent communication performance metrics transmission is closer. Therefore, the time-reversal process does
are shown in Fig. 7. Note the depth scale change in the received more than just increase SNR. It also must be suppressing multi-
signal display. The single source data was transmitted from the path. Although both the time-reversal and single source QPSK
SRA to the VRA and still shows energy diffuse with depth. transmissions had the same SNR, the PLL was able to track
only the time-reversal transmission. Both noise and ISI affect
A BER of 218/4976 and 5160/9952 was measured for BPSK
the ability of the PLL to lock onto and track phase distortions
and QPSK transmissions, respectively, with a SNR of about
[30]. Thus, it appears that the ISI that is preventing the PLL
9 dB. The PLL did not converge for the QPSK data. These se-
from tracking the single source transmissions is not present in
quences were transmitted 9 June at [Link] UTC and [Link]
the received time-reversal sequence. Note that both BPSK and
UTC. For the broadside transmissions, a BER of 91/4976 and
QPSK have the same SNR/symbol. However, QPSK has half as
4007/9952 was measured for BPSK and QPSK, respectively,
much SNR/bit. In the digital communications literature SNR/bit
with a SNR of about 21 dB. They were transmitted 9 June at often is denoted energy per bit over noise power spectral den-
[Link] UTC and [Link] UTC. The time-reversal focus ex- sity (Eb/No). However, these two metrics are the same [32,
tended in depth less than 2 m. Although the substantial multipath Chap. 7.1].
spread severely degraded the single source results, the time-re- The effective two-way channel response from the time-re-
versal process was able to compensate for the multipath yielding versal process is characterized in time by a temporal mainlobe
good results. A BER of 15/4976 and 14/4976 was measured for and multiple sidelobes (i.e., the summation term in (1) and also
BPSK and QPSK, respectively, with a SNR of about 24 dB. called the Q-function in [18]). Since time reversal implements
Accidentally, only 5 sec of QPSK was transmitted. The BPSK the matched filter of the ocean channel response, the temporal
transmission appears to have more spread in the scatter plot than sidelobes are expected to be symmetric unless the ocean fluctu-
the QPSK transmission. The channel responses were captured ates between PS reception and retransmission. These sidelobes
9 June at [Link] UTC and [Link] UTC and their corre- are a form of self-generated ISI and are shown in Fig. 9(a). The
sponding BPSK and QPSK communication sequences were re- overlap of prior and later time-reversal foci on the symbol of
transmitted at [Link] UTC and [Link] UTC, respectively. interest leads to ISI and an increase in the BER. This is not a
860 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 4, OCTOBER 2005

Fig. 9. Self-generated ISI in the time-reversal communications system.

(a) The envelope of the time-reversal focus is shown with the residual sidelobes
that lead to ISI. Since the ISI is self-generated, increasing transmit power will
not improve the BER. The dashed line represents the average self-generated
Fig. 8. Theoretical BER versus SNR/bit for additive white Gaussian noise sidelobe level taking into account overlap from previous and future symbols.
(AWGN). Ideally, the measured signals would lie upon the thick black line (b) BER for the range-dependent environment. The BER of both the original
if the signals were only subject to AWGN. ISI and self-generated sidelobes time-reversal receptions and those receptions reduced in SNR with AWGN
(in the case of time reversal) lead to higher BER. The SNR of the measured are shown. The expected BER approaches a constant in the limit of the signal
time-reversal transmission was lowered with AWGN until equal to the single power going to infinity. The dashed line represents the theoretical BER in the
source transmission. The time-reversal process does more than provide presence of ISI that is 9 dB down from the signal power.
SNR gain since it also reduces ISI. The first and second panels show the
effective time-reversal processing gain in the range independent and dependent
environments, respectively. IV. ADAPTIVE CHANNEL EQUALIZATION
In practice, the most successful coherent communication sys-
transducer characteristic effect. Increasing transmit power will
tems have utilized some form of channel equalization. Here the
not suppress this ISI because these sidelobes are signal-related.
application of an adaptive decision feedback equalizer (DFE) on
Although the SNR is 24 dB for the FAF-00 range-dependent
the time-reversal communication sequences is presented.
case, the signal power was only about 9 dB higher than the av-
erage self-generated ISI taking into account overlap from pre-
vious and future symbols. This average peak-to-sidelobe ratio is A. Receiver Stucture
depicted as a dotted line in Fig. 9(a). The range-dependent BER The receiver structure consists of a baseband equalizer and
results of the PSK sequences are shown in Fig. 9(b) with and a phase lock loop (PLL) that compensates for frequency offset
without additional noise. The dashed line represents the theoret- and jitter. The basic components of the adaptive channel equal-
ical probability of error in the presence of ISI that is 9 dB down izer are shown in Fig. 10 [3]. The equalizer and PLL are jointly
from the signal power [32]. The value of 9 dB was fitted to the optimizing the filter weights while simultaneously performing
curve and is equivalent to the effect of the self-generated overlap carrier phase recovery. The recursive least squares (RLS) algo-
observed. Underlying the theoretical probability of error is the rithm was used to adaptively perform this optimization.
assumption that the multipath can be modeled as a Gaussian The receiver structure shown in Fig. 11 is a decision-feed-
random process with zero mean in-phase and quadrature com- back equalizer (DFE). The DFE is often employed due to its
ponents. The self-generated ISI effectively becomes a part of low complexity and ability to track time-varying channels. Once
the noise term causing the BER to become constant despite in- the equalizer converges and is operating in a low BER regime
creasing signal level. In the limit of the signal power going to using a training sequence, it is placed in decision-directed mode
infinity, the predicted BER is dominated by the ISI causing it to where the previous decisions are fed back. If those decisions are
approach a constant. correct, then the DFE should be able to perfectly remove the ISI
EDELMANN et al.: UNDERWATER ACOUSTIC COMMUNICATIONS USING TIME REVERSAL 861

Fig. 10. A flow chart showing the major components of an adaptive channel equalizer. The joint phase tracking and equalization component is expanded in
Fig. 11.

Fig. 11. The decision feedback adaptive channel equalizer receiver structure.

caused by past symbols leaving only error from future symbols.

Both past and future symbols can contribute to ISI.
The DFE filter weights are optimized using a deterministic
least squares criterion. There are feedforward filter weights
operating on input symbols and feedback filter Fig. 12. (a) The response of the channel. The peak represents the strongest
weights operating on previously detected symbols. At the arrival followed by extended multipath. (b) The time-reversal, two-way channel
response. The time-reversal process suppressed temporal sidelobe amplitude
output of the equalizer, an estimate of the originally trans- though extended sidelobe structure in the negative time direction. In a static
mitted data symbol is made. The estimate is passed to a ocean, (b) would be perfectly symmetric in time.
decision device which yields where denotes a hard de-
cision. After being complex basebanded and low pass filtered
from a depth of 60 m to a receiver at a depth of 60 m over a
to remove higher frequency terms, the received signal can be
10-km propagation path. The temporal main peak is followed
written as:
by surface and bottom reflected paths. These later arrivals are
the primary cause of ISI. The channel response was estimated
(3) by correlating the measured FAF-00 single source communica-
tion sequence with the entire known data sequence, essentially
treating the data as a form of pseudo-random noise. For adap-
where are the -ary data symbols transmitted every s.
tive channel equalization to completely remove ISI, the number
The complex envelope of the channel impulse response, ,
of filter weights should span the entire channel response. In this
disperses the transmitted symbols. The signal also is corrupted
example, the filter must span about to 20 ms where time
by additive white Gaussian noise which after filtering is de-
zero is defined as the time corresponding to the primary arrival.
noted . The squared error between the true and estimated
The filter weight spacing in the receiver structure is fixed at one
data symbol is defined as . In a time-reversal commu-
filter weight per symbol. During the FAF-00 experiment, there
nication system, the channel impulse response is replaced
were 24 samples in a 2 ms symbol corresponding to a sampling
with the two-way time-reversal channel response also called the
rate 12 kHz. Thus, the input to the adaptive equalizer is the
Q-function as shown in (1).
output of a quadrature integrate-and-dump processor which av-
eraged over the 24 samples. Not described here, some systems
B. Equalizer Implementation employ fractionally-spaced equalizers that have multiple filter
This section discusses the practical aspects of implementing weights per symbol [3].
an adaptive channel equalizer. It also illustrates the advantages Fig. 12(b) depicts the two-way time-reversal channel re-
and challenges of cascading an adaptive channel equalizer with sponse. The time-reversal process has suppressed the amplitude
a time-reversal communications system. of the temporal sidelobes but it has resulted in their temporal
For an adaptive channel equalizer to completely remove ISI, elongation. The net effect is reduced ISI (Section III-C). Un-
the filter coefficients must converge and track the channel re- fortunately, the span of the equalizer now should be increased
sponse estimate in addition to adequately spanning the channel thus potentially slowing convergence.
response. While the shallow water ocean is a highly reverberant The time-reversal focus seemingly is more robust to ocean
environment, noise and attenuation result in a finite duration of fluctuations than one-way transmissions. The two range inde-
significant multipaths in practice. A channel response measured pendent channel responses versus time are shown in Fig. 13. The
4 June at [Link] UTC is shown Fig. 12(a). The PS transmitted one-way measurements shown in Fig. 13(a) were made from
862 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 4, OCTOBER 2005

Fig. 13. Measured time-evolving channel impulse responses. (a) One-way

measurements. (b) Time-reversal measurements. The time-reversal results are
more stable although the focus begins to degrade after approximately 30 min.

matched filtering 1000 Hz bandwidth 100 ms FM chirps cen-

tered at 3500 Hz transmitted from a TRM element at 60-m depth
to a VRA transducer at 60-m depth. These measurements were
made 31 May starting at [Link] UTC. The envelope of the
time-reversal foci shown in Fig. 13(b) were made on the same
day as the FM chirps starting at [Link] UTC. The stability
observed in Fig. 13 suggests that the equalization filter weights
only need to track gradual changes in the focus. In addition, the
time-reversal process mitigates channel fading and yields a high
SNR both of which are highly desirable in any single channel
adaptive equalization system.

C. Adaptive Channel Equalization Results

This section will present results of cascading a single channel,
adaptive equalizer with the received time-reversal communica-
tion sequences.
Results from cascading the RLS DFE with all of the time-
reversal communication sequences are shown in Fig. 14. The
equalizer used , and with 100 training sym-
bols. The proportional and integral tracking constants adjust the
capture and lock ranges of the embedded PLL [3]. The propor-
tional tracking constant, was set between 0.0001–0.001 and
, the integral tracking constant was an order of magnitude
less. Lastly the forgeting factor was set between 0.96–0.99.
The left hand column of Fig. 14 shows the performance of the
time-reversal communication sequence after simple phase cor-
rection as previously shown in Figs. 6 and 7. Though the results
are good, residual self-generated ISI is degrading the symbol
scatter plots. The right hand column shows the time-reversal
communication sequence after adaptive channel equalization.
In three of the four cases, the DFE visually appeared to improve
the separation of the symbol scatter clusters and lowers the BER.
The first row shows the scatter plots of the time-reversal BPSK
communications sequence before and after the DFE has tight-
ened the cluster of IQ points. For QPSK and higher constel-
lations, tightening the scatter clusters into distinct decision re- Fig. 14. The original time-reversal communication scatter and MSE plots in
the left column are compared with the RLS DFE post-processed results that
gions is essential for achieving higher bit rates. The second show tighter groupings and diminished ISI. Only for the range-dependent BPSK
row shows their corresponding MSE versus symbol number. In transmission case did the DFE experience significant difficulty converging.
EDELMANN et al.: UNDERWATER ACOUSTIC COMMUNICATIONS USING TIME REVERSAL 863

all but the range-dependent BPSK transmission shown in the [5] A. Parvulescu, “Matched-signal (’MESS’) processing by the ocean,” J.
third set of plots, the RLS DFE processing produced improved Acoust. Soc. Am., vol. 98, pp. 943–960, 1995.
[6] M. Fink, “Time reversal mirrors,” in Acoustical Imaging, B. F. Jones,
results. In this third case, the filter weights were changed to Ed. New York: Plenum, 1995, vol. 25, pp. 1–15.
, and to optimize the filter’s performance. [7] , “Time reversed acoustics,” Phys. Today, vol. 50, no. 3, pp. 34–40,
Mar. 1997.
[8] D. R. Jackson and D. R. Dowling, “Phase conjugation in underwater
V. SUMMARY acoustics,” J. Acoust. Soc. Am., vol. 89, pp. 171–181, 1991.
[9] , “Narrow-band performance in underwater acoustics,” J. Acoust.
The value of time reversal is its ability to focus sound back Soc. Am., vol. 91, pp. 3257–3277, 1992.
at the PS location without any a priori knowledge of the ocean [10] D. R. Dowling, “Phase-conjugate array focusing in a moving medium,”
environment. The time-reversal focus enables the communi- J. Acoust. Soc. Am., vol. 94, pp. 1716–1718, 1993.
[11] , “Acoustic pulse compression using passive phase conjugate pro-
cation system to minimize multipath and channel fading thus cessing,” J. Acoust. Soc. Am., vol. 95, pp. 1450–1458, 1994.
improving the SNR and reducing ISI. This is carried out with [12] W. A. Kuperman, W. S. Hodgkiss, H. C. Song, T. Akal, C. Ferla, and D.
simply the measured probe signal and does not require channel Jackson, “Phase conjugation in the ocean: Experimental demonstration
of an acoustic time-reversal mirror,” J. Acoust. Soc. Am., vol. 103, pp.
equalization algorithms to be implemented. In addition, the 25–40, 1998.
time-reversal focus appears to be robust over periods of tens [13] P. Roux, “Time-Reversal Mirrors: Application to Waveguide Propaga-
of minutes at 3500 Hz. In a time-reversal communications tion and Characterization of Fluid Flows (French),” Ph.D., University of
Paris, 1997.
system the focus decays both in range and especially in depth [14] W. A. Kuperman, W. S. Hodgkiss, H. C. Song, P. Gerstoft, P. Roux,
at positions other than that of the PS. Because of this, only a T. Akal, C. Ferla, and D. R. Jackson, “Ocean acoustic time reversal
listener at the correct location gets the benefits of suppressed mirror,” in Proc. Fourth European Conf. Underwater Acoustics, 1998,
pp. 493–498.
multipath and increased signal strength. This has two beneficial [15] G. F. Edelmann, T. Akal, W. S. Hodgkiss, S. Kim, W. A. Kuperman, and
side effects: It reduces cross-talk at other potential receiver H. C. Song, “An initial demonstration of underwater acoustic communi-
locations and confounds eavesdroppers. In addition, the sharp cations using time reversal,” IEEE J. Ocean. Eng., vol. 27, pp. 602–609,
2002.
time-reversal focus might potentially increase the effective [16] , “Underwater acoustic communications using time reversal,” in
Shannon channel capacity by enabling simultaneous communi- Oceans Hawaii MTS/IEEE Conf. and Exhibition, vol. 4, 2001, pp.
cation with multiple receivers. 2231–2235.
[17] D. Rouseff, D. Jackson, W. Fox, C. Jones, J. Ritcey, and D. R. Dowling,
This paper presented 10 sec time-reversal communication “Underwater acoustic communication by passive-phase conjugation:
sequences and compared their performance results with both Theory and experimental results,” IEEE J. Ocean. Eng., vol. 26, pp.
single source and broadside sequences. The time-reversal 821–831, 2001.
[18] T. C. Yang, “Temporal resolutions of time-reversed and passive-phase
communication transmission performance results consistently conjugation for underwater acoustic communications,” IEEE J. Ocean.
acheived lower bit error rates than the other transmission types. Eng., vol. 28, pp. 229–245, 2003.
In most cases, the time reversal transmissions are limited by [19] A. Derode, A. Tourin, and M. Fink, “Ultrasonic pulse compression with
one-bit time reversal through multiple scattering,” J. Appl. Phys., vol.
self-generated ISI resulting from residual temporal sidelobes 85, pp. 6343–6352, May 1999.
of the focus. The time-reversal communication sequences were [20] A. Baggeroer, W. A. Kuperman, and H. Schmidt, “Matched field pro-
generated 7 to 9 min after the initial PS capture at the SRA yet cessing: Source localization in correlated noise as an optimal parameter
estimation problem,” J. Acoust. Soc. Am., vol. 83, pp. 571–587, 1988.
still achieved a strong focus. [21] R. K. Brienzo and W. S. Hodgkiss, “Broadband matched-field pro-
Residual ISI and temporal degradation of the focus limit the cessing,” J. Acoust. Soc. Am., vol. 94, no. 5, pp. 2821–2831, Nov. 1993.
ability of the TRM to transmit higher order constellations over [22] S. Kim, G. F. Edelmann, W. A. Kuperman, W. S. Hodgkiss, H. C. Song,
and T. Akal, “Spatial resolution of time-reversal arrays in shallow water,”
extended periods of time. The time-reversal sequences are well- J. Acoust. Soc. Am., vol. 110, pp. 820–829, 2001.
suited for post-processing using adaptive channel equalization [23] W. S. Hodgkiss, W. A. Kuperman, H. C. Song, T. Akal, C. Ferla, and
due to their stability, high SNR, and resistance to fading. An ini- D. Jackson, “Time reversal focusing with less than a full water column
source array,” J. Acoust. Soc. Am., vol. 102, no. 5, pp. 3171–3171, 1997.
tial look at implementing a DFE to reduce the residual time-re- [24] D. F. Gringas and P. Gerstoft, “Inversion for geometric and geoacoustic
versal ISI was presented. The DFE made improvements to both parameters in shallow water: Experimental results,” J. Acoust. Soc. Am.,
the scatter and BER in all but one of the four time-reversal com- vol. 97, pp. 3589–3598, 1995.
[25] J.-P. Hermand and P. Gerstoft, “Inversion of broad-band multitone
munication sequences. acoustic data from the YELLOW SHARK summer experiments,” IEEE
J. Ocean. Eng., vol. 21, pp. 324–346, 1996.
ACKNOWLEDGMENT [26] S. R. Khosla and D. R. Dowling, “Time-reversing array retrofocusing in
noise environments,” J. Acoust. Soc. Am., vol. 109, no. 2, pp. 538–546,
The authors are grateful for the contributions of the NATO 2001.
[27] , “Scaling of time-reversing array performance in the presence of
Undersea Research Center in carrying out this joint experiment. ambient noise,” J. Acoust. Soc. Am., vol. 106, no. 4, pp. 2297–2297,
1999.
[28] W. S. Hodgkiss, G. E. Edmonds, R. A. Harris, and D. E. Ensberg, “A
REFERENCES high frequency phase conjugation array,” in Proc. Oceans 2001, 2001,
[1] D. Kilfoyle and A. Baggeroer, “The state of the art in underwater pp. 1581–1585.
acoustic telemetry,” IEEE J. Ocean. Eng., vol. 25, pp. 4–27, Jan. 2000. [29] J. G. Proakis, Digital Communications. New York: McGraw-Hill,
[2] M. Stojanovic, J. A. Capitovic, and J. G. Proakis, “Adaptive multi- 1995.
channel combining and equalization for underwater acoustic commu- [30] S. Hinedi and W. C. Lindsey, “Intersymbol interference effects on BPSK
nictions,” J. Acoust. Soc. Am., vol. 94, pp. 1621–1631, 1993. and QPSK carrier tracking loops,” IEEE Trans. Commun., vol. 38, pp.
[3] , “Phase coherent digital communications for underwater acoustic 1670–1676, Oct. 1990.
channels,” IEEE J. Ocean. Eng., vol. 19, pp. 100–111, 1994. [31] J. R. Buck, J. C. Preisig, M. Johnson, and J. Catipovic, “Single-mode ex-
[4] A. Parvulescu and C. S. Clay, “Reproducibility of signal transmissions citation in the shallow-water acoustic channel using feedback control,”
in the ocean,” Radio Elec. Eng., vol. 29, pp. 223–228, 1965. IEEE J. Ocean. Eng., vol. 22, pp. 281–291, Apr. 1997.
864 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 4, OCTOBER 2005

[32] R. E. Zeimer and W. H. Tranter, Principles of Communications. New W. S. Hodgkiss (S’68–M’75) was born in Bellefonte, PA, on August 20, 1950.
York: Wiley, 1998. He received the B.S.E.E. degree from Bucknell University, Lewisburg, PA, in
1972, and the M.S. and Ph.D. degrees in electrical engineering from Duke Uni-
versity, Durham, NC, in 1973 and 1975, respectively.
From 1975 to 1977, he worked with the Naval Ocean Systems Center, San
G. F. Edelmann received the B.S. degree in physics from the University of Diego, CA. From 1977 to 1978, he was a Faculty Member in the Electrical En-
California, San Diego, in 1997, and the Ph.D. degree from the Scripps Institution gineering Department, Bucknell University, Lewisburg, PA. Since 1978, he has
of Oceanography, San Diego, CA, in 2003. been a Member of the Faculty of the Scripps Institution of Oceanography, Uni-
While conducting the research included in this paper he was part of the Ma- versity of California, San Diego, and on the staff of the Marine Physical Labo-
rine Physical Laboratory at the Scripps Institution of Oceanography. Since 2004 ratory where currently he is the Deputy Director. His present research interests
he has been with the U.S. Naval Research Laboratory, Washington, DC. His are in the areas of signal processing, propagation modeling, and environmental
research interests include atmospheric and ocean acoustics, digital telemetry, inversions with applications of these to underwater acoustics and electromag-
sonar, infrasound, and digital signal processing. netic wave propagation.
Dr. Hodgkiss is a Fellow of the Acoustical Society of America.

H. C. Song (M’02) received the B.S. and M.S. degrees in marine engineering
and naval architecture from Seoul National University, Seoul, Korea, in 1978
and 1980, respectively, and the Ph.D. degree in ocean engineering from the W. A. Kuperman has done theoretical and experimental research in ocean
Massachusetts Institute of Technology, Cambridge, MA, in 1990. acoustics and signal processing at the Naval Research Laboratory, NATO
From 1991 to 1995, he was with Korea Ocean Research and Development Undersea Research Centre in La Spezia, Italy and the Scripps Institution of
Institute, Ansan, Korea. Since 1996, he has been a Research Scientist in the Oceanography (SIO) of the University of California, San Diego. Presently he
Marine Physical Laboratory/Scripps Institution of Oceanography, University of is a Professor at SIO and Director of its Marine Physical Laboratory.
California, San Diego. His research interests include time-reversed acoustics,
robust matched field processing, and wave propagation physics.
Dr. Song is a Fellow of the Acoustical Society of America.

T. Akal was a Principal Senior Scientist at NATO Undersea Research Center, La

Spezia, Italy, while contributing to this paper. For the past 33 years, he has been
S. Kim received the B.S. and M.S. degrees from the Seoul National University, leading research projects related to underwater acoustic and seismic propaga-
Seoul, Korea, in 1986 and 1988, respectively, and the Ph.D. degree from the tion and marine sediment acoustics. He is currently collaborating with SUASIS,
Scripps Institution of Oceanography, San Diego, CA, in 2002. TUBITAK-Technological Development Zone, Marmara Research Center, Earth
While contributing to this paper, he was part of the Marine Physical Labora- and Marine Sciences Research Institute, Gebze-Kocaeli, Turkey, as well as the
tory at the Scripps Institution of Oceanography. Since 1990 he has been with the Marine Physical Laboratory of Scripps Institution of Oceanography at Univer-
Agency for Defense Development, Korea. His research interests include ocean sity of California, San Diego, and Lamont-Doherty Earth Observatory of Co-
acoustics and sonar signal processing. lumbia University, Palisades, NY.

View publication stats