Brett E. Bissinger

Minimum Hellinger Distance based classification of underwater acoustic signals

Source classification and localization in the underwater environment is a challenging problem in part because propagation through the space- and time-varying medium introduces uncertainty in the signal. Coupled with uncertainty in environmental parameters, this uncertainty in received signals leads to a statistical treatment. Minimum Hellinger Distance (MHD) methods provide a robust and efficient framework for making classification decisions in this context.

An information theoretic performance bound for passive sonar localization of a moving source

Several passive sonar signal processing methods have previously been developed for determining the location of a source radiating tonal acoustic energy while moving through a shallow water environment. These localization algorithms rely on the complex interference pattern resulting from multipath acoustic propagation. By treating passive sonar localization as a communications problem, an information theoretic upper bound on performance can be derived. The bound is based on acoustic propagation, and depends on radial distance the source travels through the waveguide, signal to noise ratio, frequency of the radiated acoustic tone, and minimum sound speed of the problem, and resolution of the localization. An example using parameters from the SWellEx-96 experiment is shown.

Signal processing trade-offs for the quality assessment of acoustic color signatures

Acoustic color is a representation of the spectral response over aspect, typically in 2-D. The two natural axes for this representation are frequency and the aspect to the object. It is intuitive to assume finer resolution in the two dimensions would lead to more information extractable for improved quality. However, with conventional linear track data collection methods, there is an inherent trade-off between signal processing decisions and the amount of information that can be utilized without loss of quality. In this work, how the information distribution is affected with choices of representation domain will be presented. The quality metrics including resolution, signal-to-noise-ratio, and other metrics will be discussed in the context of various signal processing choices and parameters. Other representation approaches as extensions of acoustic color will also be explored, such as time-evolving acoustic color that shows how the spectral response changes within a ping cycle.

Application of a model-based depth discriminator to data from the REP11 experiment

We address application of a passive, model-based depth discriminator to data from the REP11 experiment. The method is based on a mode subspace approach (Premus, 2007) which uses environmental information along with a normal mode based acoustic simulation to predict the propagating mode structure. This mode space can be divided into subspaces representing the lower and higher order modes. Sufficient aperture yields orthogonal and linearly independent subspaces and a linear algebraic process yields orthogonalized subspaces with reduced aperture. Received data is then projected onto these subspaces and a discrimination statistic is formed. This work examines the application of this process to data from the REP11 experiment in terms of performance of the discriminator over different sets of data and levels of environmental knowledge. Work sponsored by ONR Undersea Signal Processing.

Exploiting differences in underwater acoustic signal and noise distributions to improve signal detection in low signal-to-noise ratio

Traditional models for acoustic signals and noise in underwater detection utilize assumptions about the underlying distributions of these quantities to make algorithms more analytically and computationally tractable. Easily estimated properties of the signal, like the mean amplitude or power, are then calculated and used to form predictions about the presence or absence of these signals. While appropriate for high SNR, quantities like the mean amplitude may not give reliable detection for SNR at or below 0 dB. Fluctuation based processors, utilizing additional statistics of received pressure, offer an alternative form of detection when features of the received signal beyond changes in mean amplitude are appreciably altered by the presence of a signal. An overview of fluctuation based processing will be given, with a focus on the underlying statistical phenomena that grant this method efficacy. Work sponsored by the Office of Naval Research in Undersea Signal Processing.

Correlating measured oceanographic and acoustic variability toward understanding uncertainty transfer

The Recognized Environmental Picture experiment 2011 (REP11) was conducted to support research in the areas of battlespace characterization, quantifying uncertainty, and decision support. By integrating results from these three areas, the goal is to incorporate physics based uncertainty transfer in models driving decision support tools. Generically, this is accomplished by parameterizing the environment according to need and providing the best knowledge available for each parameter including uncertainty. Experimentally, the research requires information on both the input and output sides of acoustic propagation models used for tactical prediction or decision aids. This talk describes an experimental effort to contemporaneously measure both oceanographic and acoustic quantities over a wide range of spatio-temporal scales using a combination of mobile, autonomous, and fixed assets. By measuring these quantities over the course of several days and repeating set geometries, many realizations of the oceanograp...

Modeling uncertainty in transmission loss due to spatio-temporal variation in environmental parameters

Tactical prediction and decision aid tools require acoustic propagation modeling. The effectiveness of these tools relies on knowledge of the transfer function between model inputs - either measured or predicted - and model outputs. Of particular interest is this transfer when inputs are uncertain and characterized statistically. The Recognized Environmental Picture experiment 2011 (REP11) was designed to provide observations of spatio-temporal variability in oceanographic and acoustic quantities. REP11 was comprised of multiple runs of co-located and contemporaneous oceanographic and acoustic measurements repeated over twenty-four hours. Acoustic measurements were made using a broad-band source towed along radials from a fixed receiver array. The sound speed field in the water column was sampled independently during each run using gliders, towed instruments, and moorings, each having a different spatio-temporal resolution. Due to the nature of the ocean environment, the sound speed field varied both in t...

Passive sonar source tracking using a model-based Bayesian approach

In the ocean, passive source tracking typically utilizes acoustic energy radiated by the target. Many scenarios of interest occur on the continental shelf where the water is shallow (depth is less than 200 m). Low acoustic frequencies (<1 kHz) are more useful because they suffer less attenuation due to absorption. However, low frequency acoustic signals propagating in shallow water are strongly affected by interference between multiple paths resulting from boundary interactions. These interactions cause an interference pattern in the transmission loss (TL) between the source and receiver. In this work, the pattern of TL variation has been used successfully to localize (or estimate the range and depth of) a moving source. A Bayesian localization algorithm employs knowledge of uncertainty in environmental parameters such as water column depth, sound speed profile, and bathymetry and utilizes Monte Carlo simulation to build a probability density functions (pdfs) for TL. The TL pdfs are incorporated into a recursive histogram filter as prior pdfs and used to process received signal amplitudes and generate posterior pdfs representing the probability of the source location. This paper examines how performance of the algorithm depends upon input signal-to-noise ratio.

Dependence of Bayesian passive sonar localization performance on signal‐to‐ratio.

Low frequency acoustic signals propagating in shallow water are strongly affected by interference between multiple paths resulting from boundary interactions. These interactions cause an interference pattern in the transmission loss (TL), which Jemmott [(2010)] successfully used to localize a moving source in range and depth. Jemmott’s Bayesian localization algorithm employs Monte Carlo simulations to build a probability density function (pdf) model for TL based on uncertainty in environmental parameters such as water column depth, sound speed profile, and bathymetry. The TL pdf models are incorporated into the recursive histogram filter as prior pdfs and used to process received signal amplitudes and generate a posterior pdf representing the likelihood of the source location. The localization algorithm has been shown to be robust to known uncertainty in environmental parameters, but other sources of uncertainty such as ambient noise have not been included in the work to date. This paper examines how perf...

Machine learning approach to passive sonar localization.

A machine learning‐based depth classifier for passive sonar is under development. The classifier is based on a support vector machine (SVM) paired with backward feature selection through margin‐based feature elimination (MFE). The overcomplete feature space consists of features that have been shown to work in other passive sonar classifiers, features that have potential based on efforts in similar fields, and features based on consideration of the physics of the problem. Examples include central moments, autocorrelation coefficients, and modal amplitudes. The MFE algorithm can be used to reduce the dimensionality of the feature space, ranking the features according to their utility in the classification task. The classifier is applied to a horizontal array with a source at endfire. The most powerful features are obtained using an approach similar to matched‐mode processing, except that the horizontal structure of the modes is utilized. [Work supported by Office of Naval Research Grant No. 321US.]