Antisubmarine Warfare

Recovering the 'Missing' dB
in
Active and Passive Broadband Sonar

Quiet and deadly -- who launches first? 'TMF' passive detection.

Improved acoustic signal detection in littoral regions is a major problem. The ten decibel discrepancy between the expected performance of broadband sonar systems (both passive and active) and the at-sea performance has significant implications for the effectiveness of littoral antisubmarine warfare (ASW). Recovering these "missing db" translates to an improvement of up to a factor of four, five or even greater in the range at which targets can be detected in harsh acoustic environments. Increased detection range means fewer ASW assets are needed to control a given operational area, the time required to clear the area is reduced, there is more time for decision making, and engagements are pushed further away from high value assests such as aircraft carriers.

chaotic.com has developed a class of algorithms, referred to as "Topological Matched Filters" that cost-effectively addresses these, and other, problems. In a general sense, a TMF is used to optimally separate a signal for "noise reduction" for improved signal detection, feature identification for improved classification, and feature extraction for image compression.

TECHNOLOGY

There are many approaches to signal detection, with the baseline for passive detection being a frequency-domain energy detector. For active sonar with known phase, a replica correlator is used. Under certain assumptions, these filters are known to provide optimum signal detection. However, experience has shown that because the environment is poorly modeled, the performance of all broadband sonar systems (passive and active) falls far short of optimal performance in littoral regions. Variations of these traditional linear detectors have a large installed customer and user base. Other ad-hoc methods such as neural networks require extensive (expensive) training or parameterization; may require special hardware; cannot guarantee performance metrics; and may result in sub-optimal detection in the general case because the noise/clutter models are deeply flawed (same for fuzzy logic, etc).

A comparison ('ROC' curves) of a good baseline passive matched filter detector and TMF. A ROC curve describes the trade-off between the probability of detection, the false alarm rate (fixed) and the signal-to-noise ratio required to detect a target. Up and to the left is better.

TECHNOLOGY DESCRIPTION

For active ASW, the most significant causes of poor detector performance are unrealistic and incomplete models of the true signal replica and the statistics for noise and clutter. Working with real sonar data (LBVDS) and reexamining the basic physics of underwater signal propagation in the context of TMF, we have formulated a new signal-processing model based on more realistic assumptions about noise, clutter, and reverberation than existing approaches. However, most of the observed improvement comes from mitigation of a newly identified source of mismatch in digital signal processing systems that we call "synchronization mismatch." Correcting this ubiquitous design error (caused over 30 years ago by a bad assumption in the basic sonar formulas) and implementing the optimal test statistic appropriate for nonstationary noise backgrounds results in a large gain for all environments.

Depending on the acoustic environment, our processor provides a factor of four or greater increase in the range to detect and classify targets when compared to existing active sonar signal processing. The same general approach can be used for mitigation of multipath propagation of electromagnetic signals in urban environments which has application for cell phones and other national defense communications needs.

A major improvement in active sonar performance can be achieved by repairing egregious errors that are ubiquitous in sonar signal processing.

Poorly modeled noise and clutter is the primary reason that detectors for passive signals fail to meet the theoretical limits of detection. A new algorithm, invented by chaotic.com, uses novel nonlinear filtering and signal-estimation techniques for signal separation (eg noise reduction) when the environmental clutter is poorly modeled. Using real data collected in harsh environments, detection gains of up to 6 dB over a proxy for a currently deployed system have been demonstrated for broadband signals. The algorithm that we call the "Topological Matched Filter" uses ideas from the relatively young field of nonlinear dynamics and U.S. Navy supported work on non-Gaussian signal processing.

TMF operates within the framework of traditional detection theory and is computationally cheap. Thus, it can easily be implemented in existing and planned passive ASW systems. For transient signals, the method has also demonstrated (on real contact data) an ability to optimally capture the support (beginning and end) of the signal and efficiently parse even short-duration low-SNR events -- the most difficult aspect of transient signal processing.

TMF is easily integrated into existing and planned signal-processing architectures, the only requirement being modification of the input/output protocols to match TMF to a specific application and middleware.

Download a more detailed description of our active sonar processor (pdf)


Quiet and deadly -- who launches first?	'TMF' passive detection.