Erin M. Fischell

Classification of underwater targets from autonomous underwater vehicle sampled bistatic acoustic scattered fields

One of the long term goals of autonomous underwater vehicle (AUV) minehunting is to have multiple inexpensive AUVs in a harbor autonomously classify hazards. Existing acoustic methods for target classification using AUV-based sensing, such as sidescan and synthetic aperture sonar, require an expensive payload on each outfitted vehicle and post-processing and/or image interpretation. A vehicle payload and machine learning classification methodology using bistatic angle dependence of target scattering amplitudes between a fixed acoustic source and target has been developed for onboard, fully autonomous classification with lower cost-per-vehicle. To achieve the high-quality, densely sampled three-dimensional (3D) bistatic scattering data required by this research, vehicle sampling behaviors and an acoustic payload for precision timed data acquisition with a 16 element nose array were demonstrated. 3D bistatic scattered field data were collected by an AUV around spherical and cylindrical targets insonified by a 7-9 kHz fixed source. The collected data were compared to simulated scattering models. Classification and confidence estimation were shown for the sphere versus cylinder case on the resulting real and simulated bistatic amplitude data. The final models were used for classification of simulated targets in real time in the LAMSS MOOS-IvP simulation package [M. Benjamin, H. Schmidt, P. Newman, and J. Leonard, J. Field Rob. 27, 834-875 (2010)].

Design of a general autonomy payload for low-cost AUV R&D

Advances in computer hardware and sensing technologies have enabled the design and development of low-cost autonomous underwater vehicles (AUVs). The reduced cost of new platforms such as the Bluefin SandShark present the opportunity for more frequent and high-risk AUV-based experiments and reduce the barrier to entry for multi-vehicle operations. The smaller size of these new vehicles compared to AUV platforms typically used for ocean exploration also makes deployment and operations possible in accessible smaller bodies of water such as lakes and rivers. However, the limited payload space and power availability of these platforms constrains the types of sensors that may be installed, demanding a compromise between operating range, mission duration, and sensor use. This paper describes the design of a standard payload for low-cost AUVs based on these constraints, and the implementation of that payload for autonomy, acoustic and environmental research in stand-alone configuration and on a Bluefin SandShark AUV.

Design, Implementation, and Characterization of Precision Timing for Bistatic Acoustic Data Acquisition

One of the challenges presented in using autonomous underwater vehicles (AUVs) for remote data collection is accurate time synchronization. In the case of bistatic acoustics, synchronization is required between AUV and source so that the time in which each ping is sent out is exactly known. There are three key obstacles to achieving this. First, the vehicle is submerged and therefore unable to access common time references directly. Second, the required accuracy in data acquisition far exceeds that possible using a computer clock to trigger data collection. Finally, to achieve accuracy in microseconds, the system must be characterized to eliminate delays introduced by filtering and analog-to-digital conversion. This paper describes the implementation and characterization of an accurate and precise timing and data acquisition system used on an AUV to collect acoustic array data. The timing was achieved using a combination of global positioning system (GPS) pulse-per-second (PPS) for synchronization on the surface and the Microsemi (Aliso Viejo, CA, USA) chip scale atomic clock (CSAC) for timing while submerged, with a PPS triggered data acquisition system. Characterization and calibration procedures were developed to ensure that the system met the experiment requirements, which included less than one percent of a wavelength error in phase, and one tenth of a meter accuracy in range. Analog and digital delays in the system were determined, and a method was demonstrated to further improve accuracy by dynamically estimating digital delays. The steps outlined in this paper for achieving precision data acquisition could be applied to many other remote systems that require similar microsecond accuracy.

Supervised Machine Learning for Estimation of Target Aspect Angle From Bistatic Acoustic Scattering

When an aspect-dependent target is insonified by an acoustic source, distinct features are produced in the resulting bistatic scattered field. These features change as the aspect between the source and the target is varied. This paper describes the use of these features for estimation of the target aspect angle using data collected by an autonomous underwater vehicle (AUV). An experiment was conducted in November 2014 in Massachusetts Bay to collect data using a ship-based acoustic source producing 7–9-kHz linear frequency modulation (LFM) chirps insonifying a steel pipe. The true target orientation was unknown, as the target was dropped from the ship with no rotation control. The AUV Unicorn, fitted with a 16-element nose array, was deployed in data collection behaviors around the target, and the ship was moved to create two target aspects. A support vector machine regression model was trained using simulated scattering bistatic field data. This model was then used to estimate the target aspect angle from the data collected during the experiment. The difference between the estimates was consistent with experimental observations of relative source positioning. The simulation-based model appeared successful in estimating the target aspect angle despite uncertainties in target and source location and mismatch between true environment and simulation parameters.

Characterization of underwater target geometry from autonomous underwater vehicle sampling of bistatic acoustic scattered fields

Thesis: Ph. D., Joint Program in Applied Ocean Science and Engineering (Massachusetts Institute of Technology, Department of Mechanical Engineering; and the Woods Hole Oceanographic Institution), 2015.

Estimation of the acoustic environment through machine learning techniques

The material composition of the bottom of shallow waterways can have significant effects on the corresponding acoustic environment, which autonomous underwater vehicles (AUVs) rely upon for sensing, navigation, and communication. Techniques that require an adequate environmental model are often used onboard AUVs to interpret the sensor data, but such techniques are often sensitive to even small deviations between the model and reality. A proposed approach to reduce this deviation is to use data from local soil and bathymetry surveys to generate environmental model approximations that may be loaded onto the vehicle in advance. During the early stage of deployment, the vehicle uses a K-nearest-neighbor classification approach to compare field calibration measurements with the various models, and select the most suitable solutions for use during the remainder of the active mission. Acoustic field simulations based on the environmental models are produced using normal mode theory as well as wavenumber integration, then compared with field array data. The techniques developed here could be used to facilitate the use of environment-sensitive approaches for detection and tracking during autonomous operations. [Work supported by the Office of Naval Research.]

Autonomous underwater vehicle self-localization using a tetrahedral array and passive acoustics

The recent development of very low-cost, miniature autonomous underwater vehicles (AUVs) has lowered the barrier toward the deployment of multiple AUVs for spatially distributed sensing. However, these AUVs introduce size, power, and cost constraints that prevent the use of traditional approaches for vehicle self-localization, such as Doppler velocity log (DVL)-aided inertial navigation. In this work, we describe a system that estimates the vehicles position relative to a single acoustic transmitter. The transmitter periodically outputs a linear up-chirp that is synchronously recorded by a tetrahedral ultra-short baseline (USBL) hydrophone array on the AUV. Real-time 3D phased-array beamforming and matched filtering is performed on-board the vehicle, and integrated with AUV pitch-roll-heading to calculate azimuth, inclination, and range measurements to the transmitter. Finally, a particle filter incorporates these measurements with vehicle speed estimates and a motion model to generate a positional likel...

Environmental effects on seabed object bistatic scattering classification

One of the factors that significantly affects bistatic scattering from seabed targets is bottom type. This factor has the potential to impact classification, as models that do not take bottom composition into account could improperly characterize target type, geometry, or material. This paper looks at the impact of bottom composition and self-burial on scattering from spherical and cylindrical targets in a 6.5 m deep environment with a mud and sand bottom. Sphere and cylinder scattering data from an autonomous underwater vehicle-based bistatic scattering experiment are compared to scattering simulation models with a range of bottom compositions and target burial increments. Three different sets of sediment parameters were tested. Correlation between the real and simulated data are then used to assess the similarity of each simulated scattering data set to the experiment data. Robustness to bottom composition in classification was then tested by training a model using simulated data and classifying experiment target data using a machine learning method for each environment type. Combined-environment classification models, composed of different ranges of mud depths and target burial increments, were shown to be effective at classifying the experiment data.

Passive tracking of autonomous underwater vehicles (AUVs) using a low-cost acoustic data collection system

One challenge to harbor security is monitoring and tracking autonomous underwater vehicle (AUV) activity. A self-contained low-cost acoustic data collection system (acbox) has been developed and demonstrated for this purpose. The acbox consists of an 8-element configurable off-the-shelf hydrophone array, a data acquisition system, a GPS for timing and navigation, and a computer for data logging and processing. Noise data on the Bluefin SandShark and Bluefin 21-inch AUVs were collected using the configurable array in the Charles River and Massachusetts Bay. From this experiment, AUV noise characteristics were determined. The bearing to the AUV was estimated based on beamformed and frequency filtered data, and compared to logged AUV position. The bearing estimates based on propeller noise were consistent with the reported AUV position. These results for range estimation and filtering bearing were tested to provide improved vehicle localization. Performance in the presence of boat noise was also assessed. Mo...

Multistatic acoustic characterization of seabed targets

One application for autonomous underwater vehicles (AUVs) is detecting and classifying hazardous objects on the seabed. An acoustic approach to this problem has been studied in which an acoustic source insonifies seabed target while receiving AUVs with passive sensing payloads discriminate targets based on features of the three dimensional scattered fields. The OASES-SCATT simulator was used to study how scattering data collected by mobile receivers around targets insonified by mobile sources might be used for sphere and cylinder target characterization in terms of shape, composition, and size. The impact of target geometry on these multistatic scattering fields is explored, and a discrimination approach developed in which the source and receiver circle the target with the same radial speed. The frequency components of the multistatic scattering data at different bistatic angles are used to form models for target characteristics. Data are then classified using these models. Classification accuracies were greater than 98% for shape and composition. Regression for target volume showed potential, with 90% chance of errors less than 15%. The significance of this approach is to make classification using low-cost vehicles plausible from scattering amplitudes and the relative angles between the target, source, and receiver vehicles.