In November 2012, an experiment demonstrating biological mimicry method for covert underwater acoustic communication (UAC) was conducted at Lianhua Lake in Heilongjiang China. Dolphin whistles were used for synchronization while dolphin clicks were used as information carrier. The time interval between dolphin clicks conveys the information bits. Channel estimates were obtained with matching pursuit (MP) algorithm, which is useful for sparse channel estimation. Adaptive RAKE Equalization was employed at the receiver. Bit error rates were less than 10−4 with 37 bits per second data rate in the lake trial.

Covert underwater acoustic communication is a particularly challenging task in the underwater acoustic environment. Covert communication signals ideally have a low probability of detection (LPD) and a low probability of interception (LPI). Owing to the scarce bandwidth available and the spreading phenomenon of time and frequency domains,1 it is practically impossible to obtain a consistent level of covertness at different ranges.

Yang et al.2,3 has employed direct sequence spread spectrum (DSSS) for covert underwater acoustic communication (UAC). Multicarrier orthogonal frequency division modulation (OFDM), and multicarrier spread spectrum (MC-SS) techniques have also been explored for LPI/LPD communication by many researchers.4–8 Usually, higher SNR is needed to successfully communicate messages in these schemes. As we know that the probabilities of detection and interception depend on SNR, high signal levels expose the communicating platforms. Another drawback of these schemes is their signal waveform. The signal waveforms used in these schemes are easily distinguishable by trained sonar operators due to their obvious features. These facts make the methods mentioned in the preceding text unsuitable for some special UAC situations.

Mimicry is a completely different approach for providing covert communications. Instead of reducing the SNR of a transmission to a minimum, the idea here is to design a modulation waveform that appears to be naturally occurring in the underwater acoustic environment. These may include sounds produced by dolphin, whale, cetacean, etc. The communication signal could be detected, but an adversary could easily exclude it in the process of recognition/classification. In this way, this mimicked version of the communication signal is unlikely to alert an adversary.

In this letter, we demonstrate that biological mimicry method can provide covert UAC by using intrinsic dolphin sounds. As the first step, Gray code transformation transforms the binary information from source coding into decimal delay information. Dolphin clicks modulated by these delays are used as the information carrier. In this way, the time interval between dolphin clicks conveys the information bits. Dolphin whistles are used for synchronization in this scheme. Channel equalization is performed by employing a RAKE receiver. Matching pursuit (MP) algorithm is used for channel estimation.9 The feasibility and efficiency of the proposed method have been reported by presenting results of system trials done in a lake.

Dolphin's call signals can be divided into three types. The first type of signal is a whistle signal. Dolphins use this for communication. The second type of signal is a click signal for positioning and localization. The third type of signal is emergency and analogue signal. Communication signals have durations from hundreds of milliseconds to several seconds. These are amplitude and frequency modulated pulsed signals. Energy of information in this type is concentrated in audio band. Localization signal duration ranges from tens of milliseconds to hundreds of milliseconds.10–14 

In the purposed method, whistle is used for synchronization and click is used for information transmission. Gray code transforms 6 bits of binary information from source coding into decimal delay information. Comparing direct decimal to binary conversion, using Gray code helps to reduce the errors in the delay estimation at the receiver. These errors may occur due to noise or other interruptions. With the Gray code, the adjacent time delay errors could be only 1 bit, which is an advantage.

Figures 1(a) and 1(b) indicate the frame structure of mimicked bio-signal and the time delay-coding scheme of Dolphin click. τdi (i = 1, 2, 3,…, L) is the coding time, which refers to each click signal corresponds to a click time delay difference, Tpi is the time duration of each click signal, Tc is the longest coding time. For each n bits of encoded information, the encoding time is divided into (2n1) parts. In this way, the coded quantization interval becomes Δτ=Tc/(2n1). From this the time delay can be calculated as:

τd=kΔτ,k=0,1,,2n1.
(1)
FIG. 1.

(a) Frame structure of mimic bio-signal combined with whistle and clicks. (b) Time delay coding scheme of Dolphin clicks, Tpi is the time duration of each click signal, τdi is the coding time. (c) Receiver block diagram with MP and RAKE. (d) The adaptive RAKE receiver for signal combining, tap delay is acquired from MP channel estimation in the time domain.

FIG. 1.

(a) Frame structure of mimic bio-signal combined with whistle and clicks. (b) Time delay coding scheme of Dolphin clicks, Tpi is the time duration of each click signal, τdi is the coding time. (c) Receiver block diagram with MP and RAKE. (d) The adaptive RAKE receiver for signal combining, tap delay is acquired from MP channel estimation in the time domain.

Close modal

Here k is the decimal information transformed by Gray code from binary information source. For example, if each encoded element has n = 6 bit information, the encoding time would be divided into 63 parts; if the digital information is “1 1 0 1 1 0,” then there would be k = 36.

As the width and encoding time of each click is variable, the communication data rate the system is

ν=L×log2(TcΔτ+1)/1L(Tpl+τdl).
(2)

From Eq. (2) it can be inferred that for a fixed encoding time Tc, the communication rate depends on the amount of information per encode element. The larger the amount of information per encode element n, the faster the communication rate. This reduces coded quantization interval Δτ. The reduction in quantization interval Δτ requires high system accuracy. In other words, more accurate time delay estimation is required for faster communication rate. It highlights the importance of the quantization interval Δτ in this scheme.

At the receiver, side synchronization is achieved by whistle. After synchronization, channel impulse response is estimated in the time domain. The estimated channel taps are sent to the RAKE receiver to realize the equalization with maximal-ratio combining, and the click signals are extracted. Signals after equalization process are processed to get the time delay information in the multichannel copy correlator. This delay information is finally used to get the binary information. Receiver block diagram of the system is given in Fig. 1(c) while the adaptive RAKE receiver block diagram is given in Fig. 1(d).

As correlation based processors in multipath underwater environment produce ambiguous output, and the correlation properties of dolphins' sounds are far from ideal, equalization should be performed on the received signal for accurate estimation of the time delay interval. As the signals in the proposed method are not traditional carrier signals, the traditional decision feedback equalization is not appropriate. To realize equalization, matching pursuit sparse channel estimator and adaptive RAKE receiver were used.

The MP algorithm is an iterative procedure that can sequentially identify the dominant channel taps and estimate the associated tap coefficients. At each iteration, it selects one column of S that correlates best with the approximation residual from the previous iteration. We consider the data pass through the underwater multi-path channel. The received signal samples can be expressed as

[y(0)y(1)y(N1)]y[s(0)00s(1)s(0)0s(N1)s(N1)S(NL)]S[h(0)h(1)h(l)]h+[w(0)w(1)w(N1)]W,
(3)

where s(n) is transmitted symbol, h(n) is the nth channel tap, L is the largest delay, and w(n) is complex additive white Gaussian noise (AWGN) with variance σw2. We can rewrite Eq. (3) as

y=l=0L1Slh(l),
(4)

where Sl is the column of the training matrix S. As most components of the channel tap are usually zero or very mall, the received data y can be formed as the liner combination of a small number of the columns of S.

Here, the residual vector rp1 is defined as y minus the contributions of all columns identified in the previous p − 1 iterations, with r0=y. At the pth iteration, the column of S onto which the residual vector rp1 has the maximal rank-one projection denoted as Sp is written as

sp=argmaxjIp1|Sjhrp1|2Sj2,
(5)

where Ip{s1,s2,,sp1} is the index set of all previously selected columns. Then ĥp, the element of ĥ associated with Sp, is found as follows:

ĥp=Ssphrp1Ssp2.
(6)

The residual vector is computed as

rp=rp1ĥpSsp.
(7)

There are two criteria for termination of iterations in MP algorithm. One termination criterion can be obtained by setting the total number of iterations P. The other termination criterion is to set a tolerance level ε. We can terminate when the relative duality gap falls below ε.

To evaluate the performance of the bionic underwater acoustic communication system experiments were carried out in October 2012 at Lianhua Lake in Heilongjiang, China. The experimental setup, sound speed profile and time-varying channel impulse response (CIR) estimated by a group of LFM signals is depicted in Fig. 2. Here one can see that max delay spread of the channel is about 30 ms.

FIG. 2.

(Color online) Schematic of the covert experiment conducted in the lake of China. The water depth was 20 m. The transmitter and receiving hydrophone were about 6 m below the boats with a distance of 2 km as illustrated in (a). (b) The sound speed profile measured in the experiment. (c) Time-varying channel impulse response. (d) Channel impulse response estimated by MP method using dolphin whistle.

FIG. 2.

(Color online) Schematic of the covert experiment conducted in the lake of China. The water depth was 20 m. The transmitter and receiving hydrophone were about 6 m below the boats with a distance of 2 km as illustrated in (a). (b) The sound speed profile measured in the experiment. (c) Time-varying channel impulse response. (d) Channel impulse response estimated by MP method using dolphin whistle.

Close modal

In the experiment, we sent mimicked bio-signal with a different number of clicks ranging from 5 to 10. The time interval between clicks conveys 6 bits of data. Time-delay resolution is 1 ms, which implies that the longest time delay of adjacent clicks is 32 ms. Transmitted signals were modulated according to the frame structure given in Fig. 1(b). Bit error rate (BER) and data rate are shown in Table I with different clicks in the experiment.

TABLE I.

Experimental result with different dolphin clicks.

 Data rateBER before equalizationBER after equalization
Number of clicks(bps)(%)(%)
29.2312 13.33 
29.1168 35.34 
33.5163 29.44 
36.3047 22.38 
36.7781 20.83 
10 37.1656 14.81 
 Data rateBER before equalizationBER after equalization
Number of clicks(bps)(%)(%)
29.2312 13.33 
29.1168 35.34 
33.5163 29.44 
36.3047 22.38 
36.7781 20.83 
10 37.1656 14.81 

As an example we take a group of signals that has 10 clicks. The time duration of whistle signal is 0.52 s and the duration of information sequence is 1.453 s. The communication rate is 37.16 bps. Channel is estimated by MP method. The channel estimate is sorted for 50 largest channel taps. The maximum time delay found from the channel taps is 35 ms. The estimated channel impulse response is shown in Fig. 2(d). From the diagram, there are five obvious groups of paths and the largest channel multipath time delay was about 30 ms. Correlation with both LFM and the MP method results in a similar kind of channel estimates.

Synchronization is accomplished by the whistle based on the correlation processor, and the ambiguity plot is shown in Fig. 3(b). Take one click, for example, the ambiguity function is also calculated and given in Fig. 3(d). Figure 4(a) illustrates all transmitted clicks signal. The self-correlation of each code element with itself and the cross correlation with others is shown in Fig. 4(b). Correlation results have been normalized by the maximum of correlation of code element sequence. Although we can get the correlation peak and can demodulate by finding the max peak position, the side lobes of the correlation curve are high because of the presence of the narrow band tonal signal in dolphins' calls. Figure 4(c) represents the received signal clicks at the receiver hydrophone. Compared with the transmitted signals in Fig. 4(a), the received signal has similar envelope, but its level is lower than that of the transmitted signal. Time domain waveform of the received signal after RAKE processing is shown in Fig. 4(e). Figure 4(d) shows the correlation output of the received signal before equalization. Figure 4(f) shows the correlation output of the received signal after RAKE processing. Figure 4(d) reveals that the amplitude of the correlation sequence at the output of direct multi-channel correlator is very low because of multipath and noise. If the signal is demodulated at this stage, this would result in high bit error rate, which is evaluated to be around 22.3%. After the RAKE processing and Gray code decoding, a zero bit error rate is achieved. The numerical value in Fig. 4 is the time delay estimate between adjacent clicks, including pulse-width of each click. The signal used in this scheme is a recorded marine mammal sound so it would be hard to distinguish it from other environmental sounds present in the water. In this way, the covertness in the communication scheme is established.

FIG. 3.

(Color online) Time domain waveform of whistle signal that is used for synchronization is shown in (a) while (b) is its ambiguity plot. (c) One of all the transmitted clicks. (d) The ambiguity plot of the click.

FIG. 3.

(Color online) Time domain waveform of whistle signal that is used for synchronization is shown in (a) while (b) is its ambiguity plot. (c) One of all the transmitted clicks. (d) The ambiguity plot of the click.

Close modal
FIG. 4.

(Color online) Time domain waveform and correlation output of the click sequences. (a) Transmitted dolphin clicks. (b) Correlation output of transmitted clicks. (c) Received clicks before equalization. (d) Correlation output of received clicks before equalization. (e) Received clicks after equalization. (f) Correlation output of received clicks after equalization.

FIG. 4.

(Color online) Time domain waveform and correlation output of the click sequences. (a) Transmitted dolphin clicks. (b) Correlation output of transmitted clicks. (c) Received clicks before equalization. (d) Correlation output of received clicks before equalization. (e) Received clicks after equalization. (f) Correlation output of received clicks after equalization.

Close modal

A new and novel approach for covert underwater acoustic communication using dolphin sounds has been presented. The experimental results of the proposed scheme show that the scheme is capable of achieving low data rate communication in very shallow-water environment. Compared with the traditional methods of OFDM and spread spectrum, the bionic version can provide a truly covert scheme, although the scheme is not a strictly LPD scheme but interceptor that would exclude the marine animals' voice in the threat evaluation/classification phase. The SNR of this scheme needs not to be reduced due to this fact. In this way, covert communications can be achieved to comparatively longer distances with improved reliability. The method can be incorporated as a solution in an underwater communication network for access of network nodes at unfriendly locations.

This work was supported by the National Natural Science Foundation of China under Grant No. 1127407 and National High Technology Research and Development Program of China under Grant No. 2009AA093601-2.

1.
T. C.
Yang
, “
Measurements of temporal coherence of sound transmissions through shallow water
,”
J. Acoust. Soc. Am.
120
,
2595
2614
(
2006
).
2.
T. C.
Yang
and
W. B.
Yang
, “
Low probability of detection underwater acoustic communications using direct-sequence spread spectrum
,”
J. Acoust. Soc. Am.
124
,
3632
3647
(
2008
).
3.
T. C.
Yang
and
W. B.
Yang
, “
Performance analysis of direct-sequence spread-spectrum underwater acoustic communications with low signal-to-noise-ratio input signals
,”
J. Acoust. Soc. Am.
123
,
842
855
(
2008
).
4.
J.
Ling
,
H.
He
,
J. A.
Li
,
W.
Roberts
, and
P.
Stoica
, “
Covert underwater acoustic communications
,”
J. Acoust. Soc. Am.
128
,
2898
2909
(
2010
).
5.
G.
Leus
,
P.
van Walree
,
J.
Boschma
,
C.
Fanciullacci
,
H.
Gerritsen
, and
P.
Tusoni
, “
Covert underwater communications with multiband OFDM
,” in
Oceans 2008
, Vol.
1–4
, pp.
391
398
.
6.
Z.
Hijaz
and
V. S.
Frost
, “
Exploiting OFDM systems for covert communication
,” in
Military Communications Conference
(
2010
), pp.
2149
2155
.
7.
G.
Leus
and
P. A.
van Walree
, “
Multiband OFDM for covert acoustic communications
,”
IEEE J. Sel. Area Commun.
26
,
1662
1673
(
2008
).
8.
P.
van Walree
,
E.
Sangfelt
, and
G.
Leus
, “
Multicarrier spread spectrum for covert acoustic communications
,” in
Oceans 2008
, Vol.
1–4
, pp.
264
271
.
9.
S. F.
Cotter
and
B. D.
Rao
, “
Sparse channel estimation via matching pursuit with application to equalization
,”
IEEE Trans. Commun.
50
,
374
377
(
2002
).
10.
K. T.
Clausen
,
M.
Wahlberg
,
K.
Beedholm
,
S.
Deruiter
, and
P. T.
Madsen
, “
Click communication in harbour porpoises Phocoena phocoena
,”
Bioacoustics
20
,
1
28
(
2010
).
11.
M. O.
Lammers
,
W. W. L.
Au
, and
D. L.
Herzing
, “
The broadband social acoustic signaling behavior of spinner and spotted dolphins
,”
J. Acoust. Soc. Am.
114
,
1629
1639
(
2003
).
12.
E. E.
Henderson
,
J. A.
Hildebrand
, and
M. H.
Smith
, “
Classification of behavior using vocalizations of Pacific white-sided dolphins (Lagenorhynchus obliquidens)
,”
J. Acoust. Soc. Am.
130
,
557
567
).
13.
P. Q.
Sims
,
R.
Vaughn
,
S. K.
Hung
, and
B.
Wuersig
, “
Sounds of Indo-Pacific humpback dolphins (Sousa chinensis) in West Hong Kong: A preliminary description
,”
J. Acoust. Soc. Am.
131
,
EL48
EL53
(
2012
).
14.
C.
Bazua-Duran
and
W. W. L.
Au
, “
The whistles of Hawaiian spinner dolphins
,”
J. Acoust. Soc. Am.
112
,
3064
3072
(
2002
).