Track-Before-Detect applications for unmanned systems

Abstract

The U.S. Navy has invested heavily over the last decade in the ability to sense and process electromagnetic, optical, and acoustic signals. The U.S. has the world’s most advanced sensors and unmatched computing power to process, analyze, and glean salient information for surveillance and tactical objectives. An emerging technology that offers the ability to transform how to approach the sensing and information extraction processes is a paradigm known as Track-Before-Detect (or TBD). This paradigm has numerable sonar applications for our submarine, surface, and surveillance communities.

Track-Before-Detect is an inversion of our thinking with regards to how to sense, process, and extract information. Classic methods serialize sensing, processing, and extraction processes to yield refined parameter estimates of various kinds (e.g. contact localization solutions). However, TBD begins the information extraction process with a host of hypotheses that individually extract their supporting evidence from the sensor data directly.

If current sensing, processing, and extraction processes are likened to a bottom-up approach, TBD is a top-down approach. Said another way, if current processes are Sensor- to-Solution, TBD is Solution-to-Sensor. Because of such, TBD is highly capable of very fluidly addressing many current and emerging challenges such as; 1) intra-sensor processing, 2) multi-sensor processing; 3) disparate sensor fusion; and 4) operator workload reduction.

For example, consider operator workload associated with sonar contact localization. Currently processes evolve serially; sensors are processed to obtain measurements, measurements are associated to one another, and associated measurements are used to estimate localization solutions. Given the sheer volume of sensors and sensor modalities operators can expend a significant amount of time in the abovementioned association step. Track-Before-Detect, however, can eliminate the abovementioned association step completely and seamlessly extract information concurrently from a multitude of sensor types and modalities. The key steps in TBD are starting with a solution hypothesis, mapping from the solution space to a sensor output space, and extracting the level of evidence in support of a given solution hypothesis.

MIKEL has recently demonstrated the Track-Before-Detect approach in two relevant applications; passive sonar contact localization and active sonar classification. The passive sonar localization application was recently published in the IEEE Journal of Oceanic Engineering and the active sonar classification application was a PMS 404 Unmanned Systems effort performed under the technical guidance of NUWCNPTs Code 85. Until MIKEL’s examination of TBD in the abovementioned academic journal, an implementation of TBD in a relevant Naval operational scenario had not been offered. Through both efforts TBD demonstrated a considerable simplification of information extraction and robust performance in the presence of recorded sea data realities such as sensor baffling, high contact density, signal fading, and high clutter levels. Through this submitted technical presentation, MIKEL aspires to motivate the consideration of TBD, by our esteemed engineers and scientists of the Naval Enterprise, for the most challenging current and future information extraction problems. The presentation will focus on introducing the concept of TBD, its benefits, and briefly describe MIKEL’s use of TBD.

 

Reference

T. Northardt, “Track-Before-Detect applications for unmanned systems,” Submarine Technology Symposium, Laurel, MD, May 16, 2019.