PAPERS

The acoustic detection and classification of completely and partially buried objects in the multipath environment of the coastal ocean presents a major challenge to the mine countermeasures (MCM) community. However, the rapidly emerging autonomous underwater vehicle (AUV) technology provides the opportunity of exploring entirely new sonar concepts based on mono-, bi- or multi-static configurations. For example, the medium frequency regime (1-10 kHz) with its bottom penetration advantage may be explored using large synthetic apertures, where acoustic information is accumulated over a series of sonar pings. The performance of such approaches is highly dependent on accurate platform navigation and timing, which poses a significant challenge to AUV developers, particularly because the navigation procedures are themselves dependent on the complicated multipath acoustic environment. Data from the GOATS'98 experiment have been analyzed to investigate the feasibility of combining seabed scattering data from consecutive pings of a fixed parametric source to form a bistatic synthetic aperture for target localization and imaging with an AUV based receiving platform. The paper describes different levels of bistatic processing including both incoherent and coherent beamforming and very large aperture interferometric approaches, and the associated performance tradeoffs are discussed.

CONFERENCES

Concurrent mapping and localization (CML) is a technique for unsupervised feature-based mapping of unknown environments, and is an essential tool for autonomous robots. For land robots, CML can be applied using video, laser, or acoustic sensors, while for autonomous underwater vehicles (AUVs) the only effective transducer in most situations is sonar. In the Generic Oceanographic Array Technology Sonar (GOATS) experiment series, CML was effectively demonstrated using a single AUV. A further hurdle in the full implementation of AUV minehunting is to re-acquire and identify targets of interest. Target re-acquisition allows other vehicles to be called into a target location to further investigate with adaptive sonar geometries or alternative sensor suites designed for classification. In this work, the features in the CML-generated map are extended from only spatial coordinates to include acoustic features such as spectral response. It is demonstrated that the inclusion of acoustic features aids in the global positioning within the map, although the fine positioning is still accomplished through standard CML. In addition, areas that are sparsely populated with targets, e.g., a sandy coastline, are shown to be more readily navigable using acoustic features.

Adaptive filtering is widely used in the sonar community to enhance sonar performance in the face of uncertain environmental properties. In effect, the filter adaptation serves to estimate the Green's function of the medium, thereby improving the target detection capability of the sonar. A more direct way to adapt to the environment would be to simply measure the environmental parameters during the sonar operation and use this information to adjust the steering vectors. Such an approach would be ideal for a sonar that remains in a temporally varying environment or for a sonar that moves through a range-dependent (or 2-D varying) environment. The drawbacks to this approach stem from the fact that a perturbation in the environmental parameters does not necessarily lead to a perturbation in the steering vectors. Therefore, a new environmental parameter measurement requires that the steering vectors be recomputed, which in most cases is a time- and computational resource-consuming process. In addition to the computational cost, the inherent uncertainty of the environmental sensor measurement is not captured by the direct computation of steering vectors. Both of these problems are alleviated with the application of Artificial Neural Networks (ANN) in the steering vector computation. In this work, the ANN approach is illustrated using a small sonar moving through a range-dependent environment, i.e. an autonomous underwater vehicle (AUV). High fidelity acoustic propagation models are applied in a comprehensive acoustic simulation package for AUV mission simulation and planning. The data generation is achieved with deterministic parameter fluctuations, while the AUV-borne environmental sensors are simulated as measuring the environmental properties with a known error distribution. The sonar performance for probability of target detection is compared between the fixed steering vector system, a system that incorporates the environmental measurements directly, and the neural network method that takes into account the sensor error distributions.

Despite decades of algorithm refinement and efficiency studies, the computational requirements of high fidelity acoustic field prediction models remain beyond the real-time or near real-time capabilities of modern computers. In many applications, however, there is ample time for pre-processing, so wise use of this time can reduce the on-line load of the field prediction. Artificial Neural Networks (ANNs), specifically Multilayer Perceptions (MLPs), are capable of approximating these acoustic models to any desired level of accuracy on a bounded domain. Most importantly, this highly accurate approximation of the field prediction can be produced almost instantaneously, after proper training of the MLP in the pre-processing phase. The speed of the MLP computation is easily sufficient to be included in real-time sonar systems, as well as at-sea trials. Futhermore, a MLP is not only capable of deterministic model approximation, but it can also be trained to produce a minimum-variance estimate in the presence of statistically predictable noise in model parameters such as sound speed. This inclusion of known statistics allows the MLP, in some cases, to be more accurate than deterministic modeling. In this paper, the MLP approach is demonstrated to be highly effective in several common ocean acoustic tasks, including fast field prediction and matched field inversion. Supporting numerical examples are included for both deterministic and stochastic models.

Autonomous underwater vehicles (AUVs)are being rapidly developed to carry sensors into the sea in ways that have previously not been possible. The full use of the vehicles, however, is still not near realization due to lack of the true vehicle autonomy that is promised in the label (AUV). AUVs today primarily attempt to follow as close as possible a preplanned trajectory. The key to increase the autonomy of the AUV is to provide the vehicle with a means to make decisions based on its sensors receptions. The current work examines the use of active sonar returns from man-like objects (MLOs) as a basis for sensor-based adaptive path planning, where the path planning objective is to discriminate between real mine and rocks. Once a target is detected in the mine hunting phase, the mine classification phase is initialized with a derivative cost function to emphasize signal differences and enhance classification capability. The algorithm is verified using at-sea data derived from the joint MIT/SACLANTCEN GOATS experiment and advanced acoustic simulation data using SEALAB. The mission oriented operating system (MOOS) real-time simulator is then used to test the onboard implementation of the algorithm.

Recent rapid developments in autonomous underwater vehicle (AUV) technology have provided the opportunity to explore new approaches for detecting and classifying mine-like objects. In particular, the mobility of the vehicles allows the implementation of deformable sonar geometries that can be adaptively controlled based on the local acoustic scattering statistics. In addition, distributing teams of receiver vehicles provides the potential to exploit the spatial diversity of the target scattering, which may contain important classification clues that are not apparent with monostatic or towed array sonar geometries. Preferred target scattering directions for both specular and, more importantly, elastic scattering returns can be interrogated by the adaptively controlled receiver platforms. The multi-platform approach can also lead to detection and classification algorithms that require significantly less computation than these traditional sonar techniques, and as such these algorithms are more readily implementable in real-time onboard the vehicles. In this work, a method of target classification is shown in which the 3-D scattered field is sampled by several receiver vehicles and information is extracted about the targets that clearly distinguish mines from rocks and rounded objects from oblong objects. The method is applicable to both buried and proud targets, and does not require the sub-wavelength accuracy navigation that is necessary for synthetic aperture sonar (SAS) imaging. The relaxation of the navigation accuracy requirement is critically important for two reasons. Firstly, the SAS micronavigation methods, while very effective in benign environments, are subject to catastrophic failure in strong currents or complex topological environments. Secondly, the micronavigation methods restrict the area search rate due to the need for significant aperture overlap between consecutive receptions. The proposed classification method is shown to be easily implementable in real-time, as is demonstrated both in simulations and in post-processing experimental data from the GOATS'98 experiment. The AUV onboard implementation is planned to be demonstrated in the GOATS 2002 experiment off the coast of Italy in June 2002. [Work supported by ONR and SACLANTCEN]

Buried target classification is of paramount importance in mine countermeasures (MCM) applications. Traditional methods use optimal detection techniques, e.g., monostatic synthetic aperture sonar (SAS) imaging, to classify objects by target shape. The shape of a target, however, does not necessarily indicate whether it is manmade (mine) or natural (rock), thus creating the possibility of an unacceptably high false alarm rate. Furthermore, the limited resolution available at the low frequencies required for seabed penetration may prevent the shape-based classification altogether, as the imaging wavelength is typically on the order of the target size. For these reasons, additional information about the character of the target is required. The current work investigates the use of frequency and aspect dependence of the target scattering for classification of canonical minelike targets, using a bistatic AUV-borne sonar receiver to interrogate spheres and cylinders in various burial conditions. Experimental evidence from goats98, an at-sea experiment performed off the coast of Elba Island, and supporting evidence from oases?3d models are used to demonstrate clear distinctions between these canonical targets, and to illustrate the potential of a multi-static system for rapid and robust MCM. [Work supported by ONR and SACLANT.]

Recent progress in underwater robotics and acoustic communication has led to the development of a new paradigm in ocean science and technology, the Autonomous Ocean Sampling Network (AOSN). AOSN consists of a network of fixed moorings and/or autonomous underwater vehicles (AUV), tied together by state-of-the-art acoustic communication technology. The GOATS'2000 (Generic Oceanographic Array Technology Systems) Joint Research Program is aimed toward the development of environmentally adaptive AOSN technology specifically directed toward Rapid Environmental Assessment and Mine Counter Measures in coastal environments. The research program combines theory and modeling of the 3-D environmental acoustics with experiments involving AOSN and sensor technology. [Work supported by ONR and SACLANT]

The GOATS (Generic Oceanographic Array Technology Systems) Joint Research Program explores the development of environmentally adaptive autonomous underwater vehicle technology specifically directed toward Rapid Environmental Assessment and Mine Counter Measures in coastal environments. The research program combines theory and modeling of the 3-D environmental acoustics with experiments involving AUVs and new sensor technology. As part of this effort, MIT is developing the GOATS multi-static sonar concept which uses a low frequency source on one AUV to sub-critically insonify the seabed over a wide area, while a formation of multiple AUVs are used for mapping the associated 3D scattered acoustic field in the water column. The spatial and temporal structure of the acoustic field is then processed using synthetic aperture sonar processing and dynamic pattern recognition and tracking for concurrent detection and classification. In addition to the increased coverage rate provided by the concurrent detection and classification and the improved bottom penetration and optimal excitation of structural resonances at low frequency, GOATS may improve detection of stealthy targets by capturing their inherent bi-static enhancement. [Work supported by ONR and SACLANT]

Buried target classification is of paramount importance in mine countermeasures (MCM) applications. Traditional methods use optimal detection techniques, e.g., monostatic synthetic aperture sonar (SAS) imaging, to classify objects by target shape. The shape of a target, however, does not necessarily indicate whether it is manmade (mine) or natural (rock), thus creating the possibility of an unacceptably high false alarm rate. Furthermore, the limited resolution available at the low frequencies required for seabed penetration may prevent the shape-based classification altogether, as the imaging wavelength is typically on the order of the target size. For these reasons, additional information about the character of the target is required. The current work investigates the use of frequency and aspect dependence of the target scattering for classification of canonical minelike targets, using a bistatic AUV-borne sonar receiver to interrogate spheres and cylinders in various burial conditions. Experimental evidence from goats98, an at-sea experiment performed off the coast of Elba Island, and supporting evidence from oases?3d models are used to demonstrate clear distinctions between these canonical targets, and to illustrate the potential of a multi-static system for rapid and robust MCM. [Work supported by ONR and SACLANT.]

Buried target detection and classification is of significant interest to the underwater mine countermeasures (MCM) community. In order to penetrate the seafloor, low frequency sources are required. For adequate temporal (cross-range) resolution, a wideband source is necessary. In this low frequency regime, the different scattering mechanisms behave as a strong function of frequency. Buried elastic targets, such as mines, exhibit strong modal behavior, generating responses delayed in time and occupying only a fraction of the full source frequency band. There is also much more sub-bottom interaction by the lower frequencies, and as such the statistics of the reverberation signal vary with frequency. This may impact the platform motion estimation capability in synthetic aperture sonar (SAS) applications. This paper details an algorithm for exploiting the temporal and frequency diversity of the signals for SAS target detection and classification. An example is provided using an at-sea bistatic SAS application with an AUV-borne sonar receiver.

The detection and classification of buried elastic targets presents a significant challenge to the shallow water mine countermeasures (MCM) community. In the shallow water environment, sub-critical insonification of the seabed is required for an acceptable sonar search rate. Previous experimental and modeling studies have shown that in the case of sub-critical insonification significant evanescent field components can be converted into vertically propagating field components making the plane-wave, ray-tracing approach to the wave propagation inadequate. Furthermore, there has also been empirical evidence that the modal response of the target may be stronger than the direct backscattered return due to the evanescent nature of the sub-bottom field. The GOATS98 experiment had as one of its main objectives to explore the possibility of classifying features of the 3-D acoustics scattering by buried and proud objects using bistatic configurations such as autonomous underwater vehicle (AUV) and horizontal line array (HLA). An unexpected physical phenomenon of a strong delayed flexural return of a flush buried spherical shell was observed in the AUV and the HLA data at frequencies much higher than predicted. A scenario similar to GOATS98 experiment has also been modeled using OASES-3D target modeling framework to further investigate scattering mechanisms of the flush buried spherical shell under evanescent insonification as well as to validate the modeling capability of the package. This paper presents a study of the interaction between the evanescent field and an elastic target, using simulated results with OASES-3D as well as experimental results from the GOATS'98 bistatic detection and imaging experiment.

Synthetic apertures have long been used to improve along-track resolution in radar systems. A central problem in the extension of synthetic aperture techniques to sonar has been platform motion estimation, as a result of the relatively high platform velocity to sound speed ratio. In the monostatic case, a displaced phase center antenna (DPCA) approach has been shown to effectively counteract distortions attributable to both platform and medium motion. However, bi-static imaging geometries are advantageous in the detection and classification of buried targets in the seabed. Recent advances in underwater vehicle technology have established the autonomous underwater vehicle (AUV) as a viable bi-static imaging platform in littoral environments. In order to utilize this technology fully, a robust bi-static platform motion estimation method for extending the physical aperture is required. In this paper, the monostatic DPCA approach is extended to the bi-static case for a randomly rough insonified seabed. [Work supported by ONR and SACLANTCEN]

A significant part of the GOATS'98 experiment was devoted to the demonstration and evaluation of the detection and classification capability of an AUV-borne acoustic receiver array. The unique data set from this experiment provides many signal processing challenges, arising from both the complex acoustic environment and from technology issues related to the developing AUV capabilities. The current work investigates the capability of the AUV to perform classical target detection, and examines some practical considerations of bistatic SAS for MCM applications. Different levels of bistatic processing, including both incoherent and coherent beamforming, are demonstrated and the associated performance trade-offs are discussed.

The acoustic detection and classification of completely and partially buried objects in the multipath environment of the coastal ocean presents a major challenge to the underwater acoustics community. However, the rapidly emerging autonomous underwater vehicle (AUV) technology provides the opportunity of exploring entirely new sonar concepts based on mono-, bi-, or multistatic configurations. For example, the medium frequency regime (1-10 kHz) with its bottom penetration advantage may be explored using large synthetic apertures, where acoustic information is accumulated over a series of sonar pings. The performance of such approaches is highly dependent on accurate platform navigation and timing, which poses a significant challenge to AUV developers, particularly because the navigation procedures are themselves dependent on the complicated multipath acoustic environment. Using experimental data from the GOATS'98 SACLANTCEN/MIT experiment, this paper describes an investigation into the feasibility of combining seabed scattering data from consecutive pings of a fixed parametric source to form a bistatic synthetic aperture for target localization and imaging with an AUV-based receiving platform. The paper describes different levels of bistatic processing including both incoherent and coherent beamforming and very large aperture interferometric approaches, and the associated performance trade-offs are discussed. [Work supported by ONR and SACLANT.]