Hoppa till huvudinnehållet
Till KTH:s startsida Till KTH:s startsida

Complex Acoustic Surveillance and Tracking (COAST) Project

Project description

  • Funding agency: CENIIT

  • PI: Isaac Skog

  • Duration: 2019 - onward

Historically, Sweden holds a strong position in underwater technology, with world-class submarines and autonomous underwater systems. The development of new technologies for underwater monitoring and surveillance is an important component needed to ensure that Sweden remains at the technological forefront within this area and maintains its industrial leadership and defense capabilities. During the last decades — driven by the development in sensing, computational, and communication capabilities — a range of new signal processing theories and methods for target detection and tracking, such as random finite set filters, adaptive waveform design, and track-before-detect filters, have been developed. Along the same line of research, methods for sensor management and distributed sensor fusion in sensor networks subjected to energy, computational, and communication constraints have been developed. These methods have successfully been applied in various terrestrial radar systems and sensor networks. However, applying these theories and methods in sub-surface monitoring and surveillance systems remains challenging due to the complex acoustic environment. This is especially true for systems operating in littoral waters, such as the Swedish coast. The shallow depths, surface heating and cooling, salinity changes, dense civil traffic, etc., result in an ever-changing acoustic environment, an environment to which the theories and methods must be adapted and tuned. Generally, the complex acoustic environment also limits the communication capabilities within underwater sensor networks to a point beyond that imposed on terrestrial sensor networks. This implies that the employed distributed target detection and tracking methods must function with minimal information exchange. And new communication strategies that make dual use of the communication signals, both for sensing and information exchange, are necessary.

In addition to the mentioned challenges, the problem of not knowing the exact location of the sensors within the surveillance system generally exists, which can severely degrade the system's performance. The uncertainties in the location of the sensors are caused by the need to, in many situations, quickly and quietly deploy the system. This implies that it is not feasible to apply active calibration methods to infer the location of the sensors. Neither can, due to the poor propagation of radio waves in water, standard radio-based positioning tools commonly employed in terrestrial systems, such as the Global Navigation Satellite Systems (GNSS), be used to infer the location of the sensors. Further, other sensing platforms, such as hydrophone arrays, may be towed behind a vessel and constantly change location and geometry. Therefore, new methods for passive calibration, a.k.a self-calibration, of the sensor locations must be developed.

Goal

The goal of the Complex Acoustic Surveillance and Tracking (COAST) project is to, in cooperation with the Dept. of Naval Systems & Underwater Technology, Swedish Defence Research Agency (FOI), and Saab Dynamics, research and develop methods to adapt current state-of-the-art target detection and tracking methods to the complex acoustic environment encountered by sub-surface systems. To obtain results that have a long-lasting impact and relevance to the Swedish industry and defense, experimentally driven research where theories and methods are developed, verified, and adapted based on sea trial data will constitute a core part of the project activities. 

Project description and work packages

The following three work packages (WPs) have been framed as a first step in developing and adapting the current state-of-the-art target detection and tracking methods to the complex acoustic environment encountered by subsurface systems.

WP1: Sensor Location Calibration

For a sensor network or sensor array to reliably detect and track signal sources, the location of the sensors within the system must be known with high accuracy. Although the location of the sensors is generally known with high accuracy when released at the surface, due to their drift in the water, the exact locations of the sensors when they have sunk to the bottom are unknown. Needless to say, for sensor systems towed behind vessels, the sensor locations are constantly changing and are, therefore, unknown. Thus, various active calibration methods for estimating the location of the sensors have been developed. However, in many situations, it is desirable to avoid active calibration methods due to time and cost constraints and the risk of revealing the existence of the surveillance system or disturbing marine wildlife. Therefore, methods that may be used to either track the motions of the sensors while they sink or to estimate their locations once they are on the bottom passively need to be developed.

Aim: To study the feasibility of using (i) inertial sensors to track the sensors while sinking or towed behind a vessel, and (ii) signals of opportunity, or in the case of towed system, the self-noise of the vessel, for estimating the locations of the sensors.

Results: A simultaneous localization and mapping (SLAM) method to calibrate the geometries of hydrophone arrays using the sound emitted from nearby ships has been developed. The performance of the proposed calibration method has been evaluated using data from two PASS-2447 Omnitech Electronics Inc. 56-element hydrophone arrays. Tests with four data sets have shown that array geometries in the northeast plane can be consistently estimated. Further, the calibration of the array geometries has been shown to increase the source localization accuracy significantly. The findings are presented in the journal paper: 

A video that shows the method in action can be found here: SLAM calibration video

WP2: Track-Before-Detect in Non-Stationary Disturbances

Track-before-detect is a powerful technique for simultaneous target detection and tracking within the same stochastic filtering framework. By coherently integrating the measured signals along possible target trajectories, thereby increasing the signal-to-disturbance ratio, targets can be detected and tracked at a low signal-to-disturbance ratio. Access to realistic target motion and disturbance models is fundamental to successfully using the track-before-detect technique. Realistic motion models are generally readily available for the considered targets. However, reliable disturbance models are generally not available. The disturbances include ambient noise caused by water turbulence, shipping, wave action, rainfall, etc., and reverberations due to bottom, surface, and in-volume reflections. The properties of these disturbances vary with the location, observation direction, time, etc. Therefore, although well-established models for individual disturbance sources under different environmental conditions exist, it is difficult to create generic applicable models. Further, many models capture only the temporal behavior of the disturbances, not the track-before-detect applications' important spatial behavior. Hence, new methods for online learning of disturbance models suitable for track-before-detect applications are needed.

Aim: To study and develop methods for online learning of disturbance models suitable for track-before-detect applications in underwater surveillance systems.

Results: Several methods for background modeling in passive sonar track-before-detect have been developed. These can be found in the publications 

A video that shows bearing only track-before-detect using the developed background modeling methods can be found here: Bearing-only TrBD

WP3: Sensor Management for Joint Sensing and Communication

Various sensor management strategies for selecting the optimal sonar waveforms, sensor locations, and paths for reconnaissance vessels have been proposed to optimize the detection and tracking performance of underwater surveillance systems. In distributed underwater surveillance networks, one of the main bottlenecks when implementing a detection and tracking framework is the, in general, limited bandwidth available for sensing and communication; the limited bandwidth generally implies that suboptimal distributed information fusion strategies have to be used, with a loss in detection and tracking performance as a result. One possible way to better use the available bandwidth and improve performance is to use the recently proposed idea of joint sensing and communication within the radar community. That is, the communication signals are designed to simultaneously be used for information exchange and active sensing. To maximize the benefits of the joint sensing and communication concept, new distributed sensor management and information strategies must be developed; strategies that, based upon currently available information, decide on which sources are to communicate, when to communicate, and what waveforms to use.

Aim: To research and develop sensor management methods for joint sensing and communication in underwater sensor networks.

Profilbild av Isaac Skog