On-board Underwater Glider Real-time Acoustic Environment Sensing

Transcription

1 On-board Underwater Glider Real-time Acoustic Environment Sensing A.Dassatti a, M. van der Schaar b, P.Guerrini a, S. Zaugg b, L. Houégnigan b, A.Maguer a and M.André b a NATO Undersea Research Centre Viale San Bartolomeo , La Spezia (ITALY) b Laboratori d'aplicacions Bioacústiques Universitat Politècnica de Catalunya Rambla Exposició s/n 88 Vilanova i la Geltrú, Barcelona (SPAIN) Abstract: Underwater gliders are autonomous vehicles that use small changes in their buoyancy in conjunction with wings to convert vertical motion to horizontal, and thereby propel themselves forward, with very low power consumption, through the ocean for a long period of time. Gliders typically make measurements such as temperature, conductivity (to calculate salinity), currents, chlorophyll fluorescence, optical backscatter and bottom depth. However, such a platform could be a good candidate if properly equipped with an acoustic payload to persistently monitor the underwater acoustic environment. For that reason, NURC and the Technical University of Catalonia (UPC) decided in 21 to jointly develop a glider acoustic payload that would provide the recording of two hydrophones but also, which is quite unique, provide in addition a real-time detection / classification (DC) capability. The DC capability will allow, while the glider being at-sea, to provide real-time feedback on the acoustic environment the glider is passing by, instead of only providing recording capability for postprocessing work as previously done in the past. The purpose of the paper is to describe the characteristics of the system that has been developed and additionally reports at-sea results from a deep-water WEBB glider operating in the Mediterranean Sea. Those results demonstrate the capability of the developed acoustic payload to detect and classify marine mammals in real-time within the glider. Examples of the noise generated by the glider are also presented I. INTRODUCTION Underwater gliders are autonomous vehicles that propel themselves, by changing their buoyancy, through the ocean for a long period of time. Gliders typically make measurements such as temperature, conductivity (to calculate salinity), currents, chlorophyll fluorescence, optical backscatter, bottom depth. They navigate with the help of periodic surface GPS fixes, pressure sensors, tilt sensors, and magnetic compasses. However, a platform having such long endurance characteristics could be considered as well for persistently monitoring the underwater acoustic environment if properly equipped with an acoustic payload. Such capability could be a potential solution for answering some of the issues of the EU s Marine Strategy Framework Directive (28/56/EC) which aims to improve the condition of all Europe s seas and ensure that human usage of these seas is sustainable. For that purpose, the EU community has a willingness to ensure that any human actions which is introducing energy, including underwater noise, is at levels that do not adversely affect the marine environment and therefore is looking for technical means to monitor that. For that reason, NURC and the Technical University of Catalonia (UPC) decided in 21 to jointly develop a glider acoustic payload capability leveraging on their work done in the marine mammals field since a lot of years, as illustrated in [1], [2], [3], [4],[5]. The acoustic payload provides not only the recording of two hydrophones acoustic channels but also, which is quite unique characteristic, provides a real-time detection / classification (DC) capability. The DC capability will allow, while 1

2 the glider being at-sea, to provide real-time feedback on the acoustic environment the glider is passing by, instead of only providing recording capability for post-processing work as previously done in the past. The idea would be at that stage that the glider will provide a summary of the collected events and data each time it is coming to the surface. The main challenge of this task is integrating DC capability, which is quite CPU demanding, in a heavily constrained environment; the main constraint is space that is very limited in the payload bay of a glider as well as maximum allowed payload weight. These two aspects limit the amount of installable energy and force globally a low power design. The purpose of the paper is to describe the characteristics of the system that has been developed and to show analysis of at-sea results that demonstrates the capability of the system for detecting and classifying in real-time marine mammals. II. HARDWARE DESCRIPTION The system is composed by three main components: a digital hydrophone, a main processing board and a power sub system. The system provides the following main services: - Acquisition of 2 acoustic channels with different gains sampled up to 192k Sample/s - Raw data storage - Complex real-time digital signal processing i. Acoustic hydrophone and pre-amplifier components Acoustic monitoring of marine mammals is highly demanding on the acoustic bandwidth as well as the dynamic range and is one of the main challenges to face in the design of passive acoustic system. In fact, marine mammals species vocalize at frequencies from 25 Hz for the fin whales up to more than khz for the porpoises. In terms of level of sounds emitted by marine mammals a level as high as 226dB re 1m can be achieved, as in the case of a sperm whale click, while in the mean time, deep water ambient noise in the bandwidth of interest, may vary from 85dB re 1µPa/ 2Hz (heavy shipping) to 2dB re 1µPa/ Hz at around 4kHz (sea state zero). For those reasons a high dynamic range is necessary. Based on that, the acoustic sub-system is derived from the NURC designed highly sensitive low-noise combined hydrophone-pre-amplifier that has been developed for the towed Compact Passive Acoustic Monitoring (CPAM) system [2]. The CPAM was designed and developed in the past years for concurrently detecting and localizing marine mammals. The selected hydrophone is the Neptune B/2 miniature hydrophone has been designed by Neptune to achieve the optimum combination of frequency, physical size and receive sensitivity. This has resulted in a hydrophone with a wide variety of applications ranging from marine mammal sound studies to the analysis of near field pressure patterns. It is important to stress that the elimination of metal components from the construction prevents corrosion and reduces sound field distortion and scattering. With its all-molded construction, the B/2 is completely waterproof down to a depth of 7 meters. characteristics Neptune B2 Resonant frequency (khz) 27 Flat Band (khz) 2 Hz 8 khz Sensitivity (db re 1V/µPa) -213 at 3 khz TABLE 1: Characteristics of NEPTUNE B2 hydrophone 2

3 Figure 1: Hydrophone sensitivity of the NEPTUNE B2 hydrophone with its pre-amplifier A pre-amplifier, with very low-noise characteristics was developed at NURC to combine with the Neptune B2 in order to obtain the high performance required for being able to perform high quality marine mammals monitoring. The output of the hydrophone pre-amplifier is sent to two analog processing chains with different fix gain pre-amplifiers, Variable gain amplifiers and variable high-pass filters. The signals at the output of the processing chains are converted in digital form with a stereo sigma delta 24 bits integrated circuit. The characteristics of the processing chains are given in Table 2 below: Characteristics Low gain High gain Pre-amplifier gain (db) 8 46 Variable gain (db) Variable HP filter (Hz) K 2 1 K 1 K Input noise (at 3 khz) (db re 1V/ Hz) -163 (G=2) -18 Conversion 2 channels, Sigma delta 24 bits Dynamic Range 118 Simultaneous sampling frequency (khz) 96/192 TABLE 2: Characteristics of processing chain ii. Processing Hardware The developed system is capable of simultaneously recording up to 2 acoustic channels, with 24 bits resolution, sampled up to 192 khz and has a 1TB disk capacity. It allows continuous Detection/Classification DC processing with endurance from 4 to 9 days depending if one or two channels are recorded. The processing capability is -provided by an Intel Atom Z53 processor. Special care in the design of the complete system was given to achieve low-power consumption that is a mandatory requirement for running multi-days mission. It must be noticed that this payload was designed in a first attempt to demonstrate the feasibility of performing both real-time acoustic DC processing within a glider. Improvements would be easy to be brought on the power consumption of any single component and the disk storage capacity, the limiting endurance factor if raw data recording is required. The endurance of the system could be as well 3

4 enhanced by recording part of the data once the DC processing has identified interesting events and/or just recording from time to time if emphasis for the mission is not anymore dedicated to marine mammals but more to ambient noise characterization measurements. The developed payload has been integrated as an initial stage in a deep water WEBB glider but has been designed to be compact enough to be integrated in any other available gliders. III. SOFTWARE DESCRIPTION The detection and classification algorithms are based on the software that was developed under the European Seafloor Observatory Network of Excellence (ESONET) and is currently used for real-time analysis of acoustic events from deep-sea platforms for LIDO (Listen to the Deep-Ocean Environment, [5,6]). Here, in addition to noise measurements, the detection is focused on impulsive events (e.g. cetacean sonar) and short-tonal signals (characteristic for whistles). Features from the detected events are used in a neural network for source classification. The LIDO processing system consists of several modules to analyze acoustic data. Current modules include noise measurements (according to the Marine Strategy Framework Directive: Descriptor 11), acoustic event detection, source classification, and source localization as shown in Table 3 below. Considering the limited power availability in the glider and an unknown test environment (no test recordings of the deployment area were available), it was decided, in the first stage, to only make use of noise measurement and detection functions. TABLE 3: LIDO Processing block diagram description IV. PAYLOAD INTEGRATION INTO GLIDER The payload integration into the glider has been challenging due to the small amount of space available within the glider. The acoustic payload, shown on Figure 2, has been integrated as a first test in a SLOCUM deep water glider. 4

5 Figure 2: Acoustic payload for glider The acoustic hydrophone was mounted as shown on the Figure 3 below on the stern of glider. The zoomed picture of the stern better shows the mounting of the hydrophone. The middle part of the glider is the payload in which has been integrated the acoustic payload. Figure 3: Acoustic hydrophone mounting on WEBB glider V. AT-SEA RESULTS The LIDO software was used in the following way for the processing of the glider at-sea data.. Noise measurements returned the calibrated RMS and peak levels of 22 second segments. The choice of the segment length was based on using a number of samples equal to a power-of-two for optimal computations. Calibration of the measurements was based on system sensitivity (hydrophone + gain) of -144 dbv re 1 µpa with the signal quantized between (-2, +2) V. Acoustic event detection was focused on four different bandwidths for impulses (e.g. cetacean sonar), below 5 Hz, (.5, 5) khz, (5,2) khz, (2,46) khz, and one bandwidth for short tonal signals (e.g. whistles) in the range of (3, 2) khz. The high frequency impulse detector was of most interest to the deployment for its use as a beaked whale detector. At-sea results from the deep-water glider operating in the Mediterranean Sea are described below and demonstrate the capability of the developed acoustic payload to detect and classify marine mammals in real-time within the glider. In the mean time, analysis of the noise generated by the glider is also presented. For information, the sampling frequency was set-up to 96 khz for the at-sea recording. 5

6 i. Glider noise assessment An example of glider noise is shown in Figure 4. The figure allows comparison of the glider s depth (blue - left axis) with the measured RMS level (green - right axis). High noise (saturation) was measured when the glider was at the surface, due to splashing water or the hydrophone bouncing on the surface. Noise peaks were also measured during engine operation; these are (poorly) visible as coinciding peaks with maximum depths. 2 Glider depth (blue) and measured RMS level (green; mean = 18.67, std = 5.62) Level (db re 1 μpa) Figure 4: Glider RMS noise (in green on the right axis with accompanying depth in blue on left axis) ii. High-Frequency sonar detection The output of the high frequency sonar detection is shown in Figure 5. The output of the detector was scaled between and. A high value indicates a high likelihood presence of impulses in the data segment (22 s) under analysis. At the surface there were always many impulsive signals due to the splashing water, otherwise there does not seem to be a correlation with glider activity and detector output. Especially important is that the glider self-noise in Figure 4 does not seem to have had an effect. The zoomed-in graph in Figure 6 shows clear detections during one dive while none were detected during three earlier dives. After inspection of the raw acoustic data these were confirmed to be cetacean sonar impulses. Based on the raw data alone, it is difficult to say which species was detected. The data may indicate sonar energy above 5 khz. In the deployment area this could be produced by pilot whales or Risso s dolphin, but beaked whales cannot be excluded. 2 Glider depth (blue) and high frequency impulse detections (green) Indicator Figure 5: Output of the high-frequency detector (in green on the right axis) with accompanying depth (in blue on the left axis) 6

7 2 Glider depth (blue) and high frequency impulse detections (green) Indicator Figure 6: Zoomed portion of Figure 5 to show detection of cetacean sonar between 7 and 7.5h. Outputs produced at the surface were due to splashing water iii. Short tonal detection Not many whistles were detected in the data. An example of the output from the short tonal detector is shown in Figure 7. The buzz in the same figure was not detected as it was outside of the configured detection range. Unfortunately, this detector was affected by the glider noise. An example is given in Figure 8 where the start-up of the glider engine is producing short tonal like signals that lead to high outputs of the detector. The glider produced a second type of noise not shown here that also occasionally triggered the detector. To avoid these detections, the lower frequency limit of the detector can be increased. In that case some sensitivity to dolphin whistles will be lost as displayed in Figure 7. Short Tonal Detection Buzz above configured detection frequency range 2 Glider depth (blue) and short tonal detections (green) Indicator Detected short tonal Figure 7: Spectrogram of raw data segment with short tonal detector output (green peaks in smaller figure). 7

8 Short Tonal Detection 2 Glider depth (blue) and short tonal detections (green) Indicator Glider engine startup Figure 8: Example of short tonal due to glider noise VI. SUMMARY This paper has described the first attempt by NURC and the Technical University of Catalonia (UPC) to develop within an underwater glider a real-time detection/classification capability of acoustic events. The first results from at-sea data have demonstrated the capability of the system to detect/classify marine mammals in real-time within a glider. Extensive at-sea campaigns are foreseen in the second part of 211 and will be used to better characterize the performance of the system. REFERENCES [1] Zimmer, W.M, Mark P. Johnson, Peter T. Madsen, and Peter L. Tyack Echolocation clicks of free-ranging Cuvier s beaked whales (Ziphius cavirostris) J. Acoustic. Soc. Am. 117 (6), June 25 [2] V. Grandi, P. Guerrini, S. Biagini, J. Osse, W. Zimmer, The Compact Passive Acoustic Monitor (CPAM), a tool for Marine Mammal Risk mitigation, OCEANS 21 conference, Seattle. [3] David T. Hughes, Arnold B-Nagy, Kendra L. Ryan, Jeffrey E. Haun Passive Acoustic Monitoring during the Sirena 1 Cetacean Survey and its Applications for Habitat Modeling.. MTS/IEEE Seattle Oceans 21 [4] André, M., van der Schaar, M., Zaugg, S., Mas, A., Morell, M., Solé, M., Castell, J.V., Sánchez, A. Real-time detection of beaked whale sonar signals over background noise and other acoustic events 23rd Conference of the European Cetacean Society, Workshop 4. Beaked whales and active sonar: transiting from research to mitigation, Mar 29 [5] Zaugg, S., van der Schaar, M., Houégnigan, L., Gervaise, C., André, M. Real-time acoustic classification of sperm whale clicks and shipping impulses from deep-sea observatories Applied Acoustics, vol 71, issue 11, p , Nov 21 [6] Lammers M, Brainard R, Au W, Mooney T, Wong K, An ecological acoustic recorder (EAR) for long-term monitoring of biological and anthropogenic sounds on coral reefs and other marine habitats Journal of the Acoustical Society of America 123(3): [7] Wiggins, S. M. and J. A. Hildebrand. High-frequency Acoustic Recording Package (HARP) for broad-band, long-term marine mammal monitoring. Pages International Symposium on Underwater Technology 27 and International Workshop on Scientific Use of Submarine Cables & Related Technologies 27. Institute of Electrical and Electronics Engineers, Tokyo, Japan