Combined in-air and underwater acoustic vector sensors to detect, localize, and track small noncooperative

Transcription

1 Combined in-air and underwater acoustic vector sensors to detect, localize, and track small noncooperative vessels E. Tijs 1, D.P. Cabo 2, and I. Artamonov 1 1 Microflown Maritime, 2 Microflown AVISA SYNOPSIS Small non-cooperative vessels pose a threat to society as they can be used for transporting drugs or illegal immigrants, or even be abused for terroristic attacks. Within the framework of a European ITEA project APPS, a sensor buoy is being developed that is equipped with an acoustic vector sensor both in air and underwater. By its working principle, an acoustic vector sensor has a good signal-to-noise ratio also at low frequencies that are usually of interest for detecting the engine or propeller of such a boat. Each acoustic vector sensor itself (in air or underwater) already points into the direction of the vessel. The position can be determined with a single buoy by exploiting the Doppler principle or the difference in the speed of sound in air and water or with multiple buoys. In this paper, experiences with this sensor buoy will be discussed and results of first experiments to determine the detectable range for typical go-fasts will be presented. Author s Biography E. Tijs obtained his bachelor degree in mechanical engineering in 2004 and has been working at Microflown ever since. He obtained his PhD on developing an in situ method to measure sound absorption in He is mainly involved with further developing sensors and applications for sound source localization and quantification. D.P. Cabo received the bachelor degree in Telecommunication Engineering in 2009 and the M.Sc. degree in Signal Processing in 2012 from the University of Vigo and he has been working at Microflown ever since. He received the second M.Sc. degree in Statistical Signal Processing in He is currently working on his PhD in Tonal and Harmonic Acoustic Source Localization at Microflown. Ivan Artamonov is a Research Engineer at Microflown Maritime. He received his B.Sc. in Telecommunication Systems from the Siberian State University, Novosibirsk, Russia in 2009, his M.Sc. in Electronics Engineering from the University of Bremen and TU Braunschweig, Germany in His current research focuses on developing underwater Acoustic Vector Sensors (AVS) and applications using these sensors. INTRODUCTION Microflown acoustic particle velocity sensors are being used for several industrial, defense, and security applications. These sensors are directional, and when three of them are combined with a sound pressure microphone an acoustic vector sensor is created. There are no intrinsic frequency limitations for determining direction as experienced with systems involving spatially distributed microphones. Particle velocity based vector sensors can be used to detect, localize, track, and classify a variety of sound sources and they can be installed on several platforms. The next section presents a more detailed description of existing systems and applications. In the remainder of this paper, the development of a new sensor platform type is described, i.e. a buoy based acoustic sensor system. This system contains a vector sensor in air, allowing the detection of direction of tonal sources, such as ships and go-fasts. Depending on the system configuration, different principles can be used to detect, localize, track, and classify. For a single vector sensor the Doppler effect can be exploited to locate tonal sources as they pass by. Multiple buoys equipped with vector sensors can be used as well, improving the robustness of the detection as external disturbances affect each sensor differently. Alternatively, a single vector sensor above the water might be combined in the future with an underwater sensor; exploiting the difference in speed of sound between the two media to determine range. A new type of underwater vector sensor called the Hydroflown to measure direction underwater is described. The results of a field test are presented to demonstrate the capability of detecting, localizing, and tracking go-fasts.

2 AMMS SYSTEM The AMMS system (Acoustic Multi-Mission Sensor) consists of several directional particle velocity sensors, a data acquisition and signal processing unit, wireless communication, and a windshield. Several options exist for geo referencing, powering and communication. Various sound sources can be measured such as gunshots, rockets, artillery, mortars, or engine driven platforms like helicopters, planes, UAVs, boats and ground vehicles. An AMMS is passive, cannot be jammed, requires no line of sight (darkness, fog, dust) and works under adverse weather conditions (heavy rain). As the AMMS is low size, weight and power, it can be deployed on all sorts of carrying platforms, such as on the ground, on a vehicle, helicopter, UAV, (fast) boat, or buoy. The housing type depends on the application. Fig. 1 and Fig. 2 show an AMMS system for installation on the ground and a UAV. Fig. 1. AMMS ground system. Fig. 2. Acoustic vector sensor with nose cone wind cap on the wing of a UAV. The detection range depends both upon the scenario and the type of signal processing algorithms applied. Scenario relevant parameters are the loudness of the noise source and environmental influences such as wind, waves, and nearby reflecting objects. E.g., light firearms can be detected up to 2 km, heavy caliber mortars can be detected up to 40 km and commercial airliners can be detected at 10 km cruising altitude. LOCALIZING SOUND SOURCES Combined with dedicated detection and classification algorithms, acoustic vector sensors can determine the direction of noise sources. For most applications not only direction but also location is of interest. The sound level perceived by the sensors usually is a poor indicator for the range. Instead, several other principles are used. Multiple vector sensors are often used together. The location can then be deducted by using triangulation (i.e. the point where vectors cross), time delays, Kalman filters, etc. Using multiple sensors also improves the robustness of the detection because different locations are usually affected by external influences differently. When multiple sensor nodes are triggered by the same event an actual detection is made. Although creating a spatially distributed array of vector sensors on land is often feasible, this is more difficult to realize in practice with buoys on sea. Instead, sound sources may be located from a single buoy using other processing methods. The first alternative localization method involves the Doppler effect. When a tonal source moves relative to the observer a change in frequency occurs. Provided that the speed and direction of the target are not changing rapidly, the rate of this frequency change can be used and combined with the direction of arrival estimated by the vector sensor to determine distance and location. Secondly, a vector sensor on a buoy might be combined with an underwater sensor. Although sound travels both through air and water, the speeds by which these sound waves travel differ. The time delay of both sound waves can be used to calculate the range. A regular hydrophone can be used for impulsive sound sources like exploding mines. However, for semi-continuous tonal sound sources that are moving such as go-fasts there is no defined beginning or end of a recording. An additional underwater vector sensor is then needed to determine the time delay at the moment when the same angle of arrival is measured by the vector sensor in air. The new underwater vector sensor called the Hydroflown is described in the next section. Additional underwater sensors improve the signal-to-noise because noise from external influences such as waves, wind, etc. affect the sound field under water and in air differently. Uncorrelated noises can be filtered out using signal processing.

3 HYDROFLOWN Usually, accelerometers or arrays of hydrophones are used to measure direction underwater. Low frequency sound waves tend to travel far and are therefore interesting for localization. However, at low frequencies the sensitivity of accelerometers is low and large arrays of hydrophones are needed. The Hydroflown is a new type of underwater sensor based on the Microflown particle velocity sensor in air [1-3], see Fig. 3. It is compact, directive, and sensitive at low frequencies. Fig. 3. Hydroflown underwater vector sensor. The Hydroflown sensor could be placed under a buoy, but also other mountings are possible. The small size of the sensor allows installation on other platforms, such as a surface vessel, submarine, UUV, or a bottom mounted setup. The underwater vector sensor could be used to localize surface vessels, underwater vessels, and torpedoes. Other applications could be sound emission tests (even in reverberant conditions), underwater navigation and communication, oil exploration, border control, marine life monitoring, etc. ACOUSTIC SENSOR BUOY A system containing an acoustic vector sensor in air has been developed that can be installed permanently on an existing buoy. The systems consists of a bipyramid structure, see Fig. 4. A seawater proof version of the AMMS is installed in the bottom half, the remaining electronics are placed in the top half. With a mechanical adaptor the system can be installed on the position where a navigation mark is normally placed. If required, the navigation mark itself can then be placed on top of the buoy. Fig. 4. AMMS buoy system for permanent installation.

4 Different types of wireless communication systems can be installed; e.g. an Xbee modem with a range of around 15km and an Iridium satellite modem with worldwide coverage. They transmit when there is a detection or when information about the system status is requested (battery power, sensor functionality, etc.). Currently, the solar panels already present on the buoy are used for generating power. Future versions of the buoy system might be equipped with additional solar panels on the sensor system or a rack of solar panels mounted below it for deployment in areas with little sun. Currently, a second buoy system is under development, which is a fully integrated system that can be rapidly deployed and has a battery lifetime for two to three days of continuous operation. DETECTING, LOCATING, AND TRACKING GO-FASTS A test has been performed to demonstrate the detection capability of surface vessels with the AMMS vector sensor system. A small boat was tracked using two sensors on a pier, see Fig. 5. Fig. 5. An AMMS system (AVS sensor, battery, and antenna) on a pier and the boat next to the pier. Detection One of the most challenging tasks for continuous acoustic source localization is the detection of weak periodic signals under low-snr conditions and non-gaussian noise [4]. Several approaches have been proposed and used. In [5], higher order statistics were used to detect and estimate the parameters of deterministic signals under Gaussian and non-gaussian noise. Sadler et al. proposed using singular value decomposition of a cumulant matrix to detect a sinusoid under non-gaussian noise [6]. In [4] stochastic resonators are used and analyzed to increase the detection probability while reducing the probability of false alarm. In this investigation, stochastic resonators combined with adaptive matched filtering techniques are used for detecting the boat s noise [4]. DOA finding Direction-of-Arrival (DOA) estimation of far-field narrowband signals has been of interest in the past few decades. Many advanced approaches have been proposed in order to get acoustic directional information using an array of omnidirectional transducers (microphones). Several well-known existing nonparametric methods are beamforming [7], Capon's method [8], and subspace based methods such as MUSIC [9]. However, such an array become large and difficult to handle when trying to cover low frequencies. In addition, the spatial coherence between transducers decreases as the size of the array increases [10], specifically for high frequencies [12]. Attempts to cover lower frequencies by increasing the spacing between transducers will lead to a reduction of the performance of the system at higher frequencies due to aliasing [11, 12] and the lack of cross-correlation or coherence between receivers [10]. Furthermore, obtaining a microphone array geometry suitable for all potential targets is challenging because different targets have different frequency ranges. The vector systems used in this research do not have these disadvantages because the particle velocity sensors themselves are directional already for all frequencies.

5 Measurement results The signals of the vector sensors in the tests with the small boat have been recorded and processed. Fig. 6 shows an example of measured signals without filtering. It shows the spectrogram of two sensors while a boat is 315 meters away from the first sensor and 350 meters from the other sensor. The SNR (the ratio of the power of the signal and the power of the noise for all frequencies) is -16dB for the first sensor and -19.5dB for the second one. Fig. 6. Spectrogram of unfiltered signals. Top: sensor one with a boat at 315 meter. Bottom: sensor two with a boat at 350 meter. Fig. 7 shows the spectrogram of the same signals after filtering. Now the signals from other noise sources (selfnoise of the sensor, wind noise, background noise, etc.) has been removed the signature of the boat can be recognize much clearer. Fig. 7. Spectrogram after filtering. Top: sensor one with a boat at 315 meter. Bottom: sensor two with a boat at 350 meter.

6 Fig. 8 presents the estimated direction of arrival on a map pointing towards the actual GPS position of the boat. For certain situations, such as for a moving source, the accuracy can even be improved by e.g. using a Kalman filter. Fig. 9 shows the directly measured direction with one sensor and estimated direction with the Kalman filter. In this case, the spacing between sensor and the boat increased from 170 meter to 350 meter. The accuracy of localization improves if multiple sensor nodes are included in the calculation. Fig. 8. Measured DOA at 350 meters from the second sensor. DOA [degrees] actual DOA measured DOA Kalman filter DOA estimate Frame [k] Fig. 9. DOA estimation with a Kalman filter. The SNR usually deteriorates if the distance to the source is increased. Still the signature of the boat can clearly be recognized when the boat is at 466 meters from the first sensor and 500 meters from the second sensor, see Fig. 10. In this situation the actual SNR of both sensors was -25 db and -26 db, respectively. Notably, this particular boat was rather silent and larger ranges of several kilometers are feasible for louder vessels.

7 Fig. 10. Spectrogram after filtering. Top: sensor one with a boat at 466 meter. Bottom: sensor two with a boat at 500 meter. CONCLUSIONS To detect maritime threats through a passive solution a sensor buoy solution is developed involving an acoustic vector sensor in air and, in the near future, underwater. The AMMS sensor used in air is sensitive also at low frequencies, which is useful for long range detection. It is directional for all frequencies and there are no sensor spacing limitations as usually experienced with arrays of microphones only. Localization and tracking can be done with multiple sensors or using a single sensor by using the Doppler frequency shift of a moving source. Alternatively, an underwater sensor can be used in which the difference in speed of sound between air and water is used to determine the range. A new type of underwater vector sensor called the Hydroflown is described, which is based on similar principles and has similar advantages as the in air AVS. With the buoy sensor presented, a variety of sound sources can be measured such as gunshots, rockets, artillery, mortars, or engine driven platforms like helicopters, planes, UAVs, and ground vehicles. In this paper, the results of a test with a boat have been shown. A good detection of this rather silent boat could be achieved even at 500 meters. Longer ranges of several kilometers are feasible for louder sound sources such as go-fasts. REFERENCES 1. M. Hezemans, K. B. Smith, J. Adeff, H.E. de Bree, Feasibility study to adapt the Microflown directional sensor for underwater use, Proc of Conf of Underwater Acoustics (UA2013), E. Tijs, H.E. de Bree, T. Akal, Sea-bottom mounted trials with Hydroflown sensors, Proc of Conf of Underwater Acoustics (UA2013), H.E. de Bree, E. Tijs, T. Akal, The Hydroflown: MEMS-Based Underwater Acoustical Particle Velocity Sensor, Proc of Conf of European Conference on Underwater Acoustics (ECUA), V.N. Hari, G.V.Anand, A.B.Premkumar, A.S.Madhukumar, Design and performance analysis of a signal detector based on suprathreshold stochastic resonance. Signal Processing Vol. 92, pp , 2012

8 5 Brian M. Sadler, Georgios B. Giannakis, Keh-Shin Lii, Estimation and Detection in NonGaussian Noise Using Higher Order Statistics. IEEE Transactions on Signal Processing, Vol. 42, NO. 10, October Estimating The Number Of Sinusoids In Non-Gaussian Noise Using Cumulants. ICASSP-91, Acoustics, Speech, and Signal Processing, Vol. 5, pp , April Henry Cox, Robert M. Zeskind, Mark M. Owen, Robust Adaptive Beamforming. IEEE Transactions on Acoustics, Speech, And Signal Processing, Vol. ASSP-35, No. 10, October H.W. Chen, J.W. Zhao, Wideband MVDR beamforming for acoustic vector sensor linear array. IEEE Proc.- Radar Sonar Navig., Vol. 151, No. 3, June R.O. Schmidt, Multiple emitter location and signal parameter estimation. IEEE Transactions on Antennas and Propagation, Vol. 34, No. 3, pp , Mar Brandstein M., Ward, D., Microphone arrays, Signal Processing Techniques and Applications. (Eds.) Springer, Wind J., de Bree H. E., Buye Xu, 3D Sound Source Localization and Sound Mapping using P-U Sensor Array, 16th AIAA/CEAS Aeroacustics Conference, Stockholm Wen-Qian Jing, Daniel Fernandez Comesan ã, David P. Cabo Sound source localisation using a single acoustic vector sensor and multichannel microphone phased arrays. Inter-noise, Melbourne, November 2014