Acoustic ship signature measurements by cross ...

Acoustic ship signature measurements by cross-correlation method Laurent Fillinger,a) Alexander Sutin, and Alexander Sedunov Stevens Institute of Technology, Castle Point on Hudson, Hoboken, New Jersey 07060

(Received 6 November 2009; revised 18 February 2010; accepted 23 February 2010) Cross-correlation methods were applied for the estimation of the power spectral density and modulation spectrum of underwater noise generated by moving vessels. The cross-correlation of the signal from two hydrophones allows the separation of vessel acoustic signatures in a busy estuary. Experimental data recorded in the Hudson River are used for demonstration that cross-correlation method measured the same ship noise and ship noise modulation spectra as conventional methods. The cross-correlation method was then applied for the separation of the acoustic signatures of two ships present simultaneously. Presented methods can be useful for ship traffic monitoring and small C 2011 Acoustical Society of America. ship classification, even in noisy harbor environments. V [DOI: 10.1121/1.3365315] PACS number(s): 43.60.Jn [NPC]

Pages: 774–778

Passive acoustic methods based on the recording and analysis of the ship’s acoustic signatures can be used for ship detection, tracking, and classification. Underwater ship noise is radiated from various parts of the vessels (machinery, pro-pulsor system, etc.)1–3 and is a combination of a continuous spectrum and of some discrete frequency lines. Even though that noise is directional2 and varies with the speed and the maneuvering3 of the vessel, it can be measured and used for ship detection and classification. The propeller rotation modulates the high frequency ship noise,4,5 and the extraction of that modulation provides another useful tool for ship detection and tracking. The ship noise is usually measured using omnidirectional hydrophones that do not allow the separation of signals from several ships. Detection, tracking, and classification of small boats in busy urban environments require methods of ship localization and separation of ship’s individual signatures. Application of acoustic arrays for these purposes requires systems with a large aperture that are expensive and difficult to operate. Much less expensive and simpler are methods based on acoustic signal recording by several hydrophones and application of generalized correlation of broad band signals. These methods were initially developed for passive submarine detection and localization,6–8 and we present here the application of cross-correlation methods for separation of acoustic signals from several boats, measurements of their acoustic signatures, and modulation spectra. This paper extends some results presented in a lay language paper.9 II. APPLICATION OF CROSS-CORRELATION FOR POWER SPECTRUM MEASUREMENTS

Let us consider how the measurements of cross-correlation can be used for estimation of the spectral density of the acoustic signal recorded by two hydrophones. a)

Author to whom correspondence should be addressed. Electronic mail: [email protected]

774

J. Acoust. Soc. Am. 129 (2), February 2011

Let us consider the signals received by two hydrophones separated by a distance L. They record the noise radiated by a ship whose direction makes an angle a with the normal to the line between the hydrophones. The distance between the ship and the hydrophones is much larger than L. An example of such a configuration is illustrated in Fig. 1. The noise radiated from the ship reaches the two hydrophones with a delay DT DT ¼

L sin a ; c

(1)

where c is the speed of sound in water. Let us assume that a single ship contributes to the acoustic field, and that the signals h1(t) and h2(t) recorded by the two hydrophones are delayed and scaled versions of the same signal h2 ðtÞ ¼ bh1 ðt DTÞ;

(2)

where b is a scaling factor (accounting for different hydrophone sensitivity and attenuation) and DT is the delay introduced in Eq. (1). The cross-correlation R12(s) of the signals h1(t) and h2(t) is defined as R12 ðsÞ ¼

ð1

h1 ðt0 Þh2 ðs t0 Þdt0 :

(3)

1

For two delayed signals of the form (2) R12 ðsÞ ¼ bR11 ðs DTÞ:

(4)

The cross-correlation R12(s) of the signals from the two hydrophones has the same shape as the autocorrelation R11(s) of the signal from the hydrophone 1, scaled by factor b, and has translated by s ¼ DT. Because the autocorrelation of a signal has maximum at s ¼ 0, the cross-correlation R12(s) has maximum at s ¼ DT. The location of the maximum of the cross-correlation can be used to estimate the direction to the ship [using Eq. (1)]. The spectrum of the cross-correlation can be used to extract the spectrum of the ship generated noise.

0001-4966/2011/129(2)/774/5/$30.00

C 2011 Acoustical Society of America V

Author's complimentary copy

I. INTRODUCTION

R12 ðsÞ ¼

N X

bi Rii ðs DTi Þ;

(8)

i¼1

where bi ¼ b1ib2i, DTi ¼ t2i t1i, and Rzi,zi(s) is the autocorrelation of signal zi(t). In Eq. (8), each ship provides a contribution whose maximum is at s ¼ DTi. If these contributions are separated by more than the width of the autocorrelation function (Ds), they can be separated. From Eq. (1), this condition can be written as FIG. 1. The noise radiated by a source reaches the sensors separated by distance L with a delay that depends on the direction a of the source with respect to the sensors.

The power spectral density S1( f ) of a signal h1(t) is defined as the Fourier transform (TF) of its autocorrelation (5)

The Fourier transform of the cross-correlation is called cross-spectral density S12 ðf Þ ¼ TF½R12 ðsÞ ¼ b expð2jpf DTÞS1 ðf Þ:

N X

b1i zi ðt t1i Þ;

i¼1

h2 ðtÞ ¼ n2 ðtÞ þ

N X

b2i zi ðt t2i Þ;

(7)

i¼1

where b1i and t1i (b2i and t2i) accounts for the attenuation and propagation time from the ship i to the hydrophone 1 (2). In such a case, the power spectral density of either hydrophone signal includes contributions from each individual ship and from the ambient noise. Since these ship sounds overlap in time, their contributions cannot be separated directly in the power spectral density recorded by a single hydrophone. Under certain assumptions, the cross-correlation offers an opportunity for separating individual ship signals. Cross-correlation of the signals Eq. (7) consists of many components. Assuming that the ambient noise on any hydrophone is correlated neither to the background noise on the other hydrophone nor to the sounds radiated by the ships, contributions to the cross-correlation involving ambient noise vanish. Furthermore, assuming that the sounds from the various ships are also uncorrelated, the only contribution that remains are the auto-correlations of the sound of the ships J. Acoust. Soc. Am., Vol. 129, No. 2, February 2011

This expression shows that if the directions of the various ships with respect to the hydrophone system are so that their contributions to cross-correlation do not overlap, they can be separated. In such a case, an individual ship signature can be estimated by taking the Fourier transform of the associated contributions in the cross-correlation in the time window around the peak of the cross-correlation function. III. RESULTS OF MEASUREMENTS

(6)

The cross-spectral density S12( f ) is equal to the power spectral density S1( f ) multiplied by coefficient b exp(2jpfDT) whose magnitude is b. Consequently, the signature of the ship (frequency dependence of the radiated noise) can be estimated independently from the signal of a single hydrophone using the auto-correlation or from the signals of two hydrophones using the cross-correlation. Lets us now consider the signals recorded on the hydrophones as the superposition of the noise zi(t) radiated from N ships plus uncorrelated ambient noise n1(t) and n2(t) h1 ðtÞ ¼ n1 ðtÞ þ

(9)

The experimental data used in that paper were collected on August 21, 2008 in the Hudson River near Manhattan by the Stevens Maritime Security Laboratory (MSL).10 ITC 6050 hydrophones produced by International Transducer Corporation were used due to their high sensitivity and low noise level in the frequency band up to 100 kHz. They were placed on the river bottom on stands with a height of 60 cm. The distance between hydrophones was about 15 m and the depth in the place of the hydrophones deployment was about 3 m. All deployed hydrophones were connected by cable to the on-board computer for data processing and storage. The signals from the hydrophones were amplified and filtered in the frequency band 5–95 kHz. This filtering was applied for suppression of the high acoustic noise level in the low frequency band, which limits the dynamic range of measurements and for elimination of spurious aliasing signals produced by electromagnetic noise at frequencies above 100 kHz. The amplified and filtered signals were digitized at a sampling frequency of 200 kHz. The boat computer was wirelessly connected with MSL Visualization and Analysis Center (VAC) for real-time display, thus allowing scientists in the VAC to control the experiments. In addition to real-time data feeds into the VAC, six video cameras have been deployed to provide real-time visual observation of experiments, as well to provide video data and to analyze water traffic. The video recording allowed for a connection of the recoded acoustic signals with a definite ship. An animation showing simultaneous video and acoustic processing is provided with the online version of the paper. Figure 2 is a snapshot of that video (see Supplemental Material11). The lower part is a floating chart representing the cross-correlation as a function of time. The time variations in the cross-correlation function are presented in the form of a floated chart similar to a spectrogram, a graph with two geometric dimensions: the horizontal axis represents time, the vertical axis is the delay between two hydrophone signals; a third dimension indicates the amplitude of the cross-correlation function. The curved Fillinger et al.: Ship signature measurements by cross-correlation

775


S1 ðf Þ ¼ TF½R11 ðsÞ:

DTi DTj ¼ L sin ai sin aj > Ds for all i 6¼ j: c

FIG. 2. (Color online) Screenshot of the video (see Supplemental Material in Ref. 11) showing data from two video cameras and floating chart of crosscorrelation [horizontal axis is time and vertical axis is time shift between two signals; see Eq. (3) in ms].

776

J. Acoust. Soc. Am., Vol. 129, No. 2, February 2011

The cross-correlation from signals acquired at a later time and presented in Fig. 4 clearly exhibits contributions associated with two different ships presented in the video (see Supplemental Material11). The contribution associated with the peak at 0.67 ms in the cross-correlation corresponds to the same ship as that presented in Fig. 3. Some difference in spectra can be attributed to the directivity of the radiated noise and to the different propagation paths in

FIG. 3. (Color online) (a) Cross-correlation showing the contribution from a ship. (b) Power spectral density estimated from the signal from a hydrophone (thin line) and from part of the cross-correlation (thick line). Fillinger et al.: Ship signature measurements by cross-correlation


line is the track of the passing ferry that is seen in the top right. Another track is visible on the cross-correlation chart that corresponds to another ship that is not yet in the field of view of the cameras. In order to validate and illustrate the potential of the proposed cross-correlation method for estimation of ship signatures, two examples are considered. The first example demonstrates that in the case of a single ship, the crosscorrelation method yields the same result as the conventional method. The second example considers a case with two ships in order to demonstrate the ability to separate the contributions from each ship. Figure 3(a) shows the cross-correlation of the acoustic signal generated by the ferry shown in the inset (the same as in Fig. 2 and in the video11). The cross-correlation was calculated using a time window of 0.66 s at a time when the acoustic signal from the ferry was much higher than the ambient noise and the signals from other boats. The contribution to the cross-correlation from this ferry is shown with a thick line in the Fig. 3(a). That contribution has been used to determine the power spectral density that is shown with a thick line in Fig. 3(b). The power spectral density estimated using the direct signal from one hydrophone is shown with a thin line and matches very well with the dependence obtained using the cross-correlation. Some difference in the frequency band above 25 kHz is probably connected with a higher ambient noise contribution to the single hydrophone spectrum. The cross-correlation method improves the signal/noise ratio and decreases the noise contribution to the measured spectrum.

FIG. 6. (Color online) Modulation spectrum of the ship of Fig. 3 computed in the same time frame using (a) cross-correlation and (b) DEMON.

shallow waters. The spectra of these two contributions are shown along with the power spectral density computed from the signal from a single hydrophone. They offer a decomposition of the spectrum measured by a single hydrophone. IV. MODULATION SPECTRUM

The other widely used method of ship classification is based on measurements of the high frequency ship noise modulation. Indeed, the noise radiated by a ship is modulated at a rate dictated by some parameters of the propeller (number of blades, rotational speed, etc.). Evaluation of that modulation provides information on that ship, such as the shaft rotation frequency, which can be used for classification. The method for estimation of the envelop modulation is known as Detection of Envelope MOdulation on Noise (DEMON)4,5 and a block diagram of this method is shown in Fig. 5(a). Since the peak of cross-correlation is proportional to the intensity of the acoustical signal, the modulation of acoustic signal intensity can recovered from the cross-correlation. For measurements of this modulation the cross-correlation has to be computed at a sufficient rate (at least twice more than the maximal modulation frequency), which also requires a short enough time window. The block diagram for that method is shown in Fig. 5(b). Figure 6 has been obtained by computing the cross-correlation using 20 ms windows at a rate of 400 Hz. For each cross-correlation, the ship contribution is extracted and its energy computed (rms), leading to one point on the envelope of the cross-correlation. The spectrum of that envelope gives

FIG. 5. Block diagram for the evaluation of the modulation spectrum using (a) DEMON and (b) cross-correlation. J. Acoust. Soc. Am., Vol. 129, No. 2, February 2011

V. CONCLUSION

It was demonstrated that cross-correlation techniques applied to ship noise recorded by two hydrophones allows the determination of the acoustic ship signature that includes

FIG. 7. (Color online) Modulation spectra computed by using cross-correlation method for two boats presented in the video (see Supplemental Material in Ref. 11) and in Figs. 3 and 4: (a) ferry, (b) fast boat, and (c) DEMON calculation of the recorded signal where the modulation of two boats are mixed. Fillinger et al.: Ship signature measurements by cross-correlation

777


FIG. 4. (Color online) (a) Cross-correlation showing contributions from two ships. (b) Power spectral density estimated from the signal from a hydrophone (thin line) and from part of the cross-correlation.

the modulation spectrum. An example is shown in Fig. 6 along with the results obtained using the conventional DEMON method. In both cases, the digitized signals were filtered in the 20–90 kHz range, in order to remove the low frequency noise that is strong in the Hudson River. The match between the two methods is very good. Figure 7 further illustrates the potential of the method. The spectrum in Fig. 7(a) corresponds to the same ship as that of Fig. 6. Both spectra show the same first and second harmonics (11 and 22 Hz) but the level of the third and higher harmonics is much smaller at Fig. 7(a). The difference may be attributed to the directivity of the radiated noise and to the different propagation paths in the shallow waters. The spectrum in Fig. 7(b) is associated with the fast boat. The frequency lines of these two spectra are present in the DEMON spectrum in Fig. 7(c) computed from the signal recorded on a single hydrophone. The presented example demonstrates the ability of the method to decompose the spectrum obtained using the DEMON method into two contributions.

ACKNOWLEDGMENT

This work was partially supported by ONR Project No. N00014-05-1-0632: Navy Force Protection Technology Assessment Project and by the U.S. Department of Homeland Security under Grant No. 2008-ST-061-ML0002. The view and conclusions containing in this document are those of the authors and should not be interrelated as necessarily representing the official policies, either expressed or implied, of the U.S. Department of Homeland Security and Office of Naval Research. 1

R. W. Fischer and N. A. Brown, “Factors affecting the underwater noise of commercial vessels operating in environmentally sensitive areas,” in Proceedings of the MTS/IEEE OCEANS 2005 (2005), Vol. 3, pp. 1982–1988. 2 P. T. Arveson and D. J. Vendittis, “Radiated noise characteristics of a modern cargo ship,” J. Acoust. Soc. Am. 107, 118–129 (2000).

778

J. Acoust. Soc. Am., Vol. 129, No. 2, February 2011

3

M. V. Trevorrow, B. Vasiliev, and S. Vagle, “Directionality and maneuvering effects on a surface ship underwater acoustic signature,” J. Acoust. Soc. Am. 124, 767–778 (2008). 4 A. A. Kudryavtsev, K. P. Luginets, and A. I. Mashoshin, “Amplitude modulation of underwater noise produced by seagoing vessels,” Acoust. Phys. 49, 184–188 (2003). 5 A. Kummert, “Fuzzy technology implemented in sonar systems,” IEEE J. Ocean. Eng. 18, 483–490 (1993). 6 B. Xerri, J.-F. Cavassilas, and B. Borloz, “Passive tracking in underwater acoustic,” Signal Process. 82, 1067–1085 (2002). 7 S. A. Stotts, J. L. Martin, and N. R. Bedford, “Multiple-source localization using GPS technology and received arrival time structure analysis in an air-deployed system,” IEEE J. Ocean. Eng. 22, 576–582 (1997). 8 R. E. Zarnich, “A fresh look at broadband passive sonar processing,” in Proceedings of the 1999 Adaptive Sensor Array Processing Workshop (ASAP ‘99), MIT Lincoln Laboratory, Lexington, MA, (1999), pp. 99– 104. 9 L. Fillinger, A. Sutin, and A. Sedunov, “Cross-correlation of ship noise for water traffic monitoring,” Acoustical Society of America 158th Meeting Lay Language Papers, http://www.acoustics.org/press/158th/fillinger.htm (Last viewed 2/17/2010). 10 B. J. Bunin, A. Sutin, and M. S. Bruno, “Maritime security laboratory for maritime security research,” Proc. SPIE 6540, 65400S (2007). 11 See supplemental material at http://dx.doi.org/10.1121/1.3365315 Document No. E-JASMAN-023005 for video showing acoustic cross-correlation processing of the received signals with simultaneous video recording of a passing boats. The upper part shows video of the Hudson River. The lower part is a floating chart representing the cross-correlation as a function of time. The time variations in the cross-correlation function are presented in the form of a floated chart similar to a spectrogram: the horizontal axis represents time, the vertical axis is the delay between two hydrophone signals; a color indicates the amplitude of the cross-correlation function. The curved lines are the acoustic tracks of the passing boats. For more information, see http://www.aip.org/pubservs/epaps.html.

Fillinger et al.: Ship signature measurements by cross-correlation


ship noise spectrum and its modulation spectrum. This technique can be applied for the separation of the acoustic signature of a single ship in presence of multiple sources, a situation in which conventional methods fails. The use of the cross-correlation presents the ability to selectively extract the acoustic signatures in the presence of multiple sources, which is especially interesting for ship tracking and identification in busy environment such as harbors. The technique can be applied with a simple and inexpensive system using just two hydrophones.