Feature Articles—November 2009 Issue
A Multitarget Tracking Method For Broadband Passive SonarMultihypothesis Tracking Concepts Could Significantly Reduce The Manual Input Needed
For Tracking Systems
By Kevin Brinkmann
Senior Systems Engineer
and
Jörg Hurka
Senior Systems Engineer
Submarine Systems Division
Atlas Elektronik GmbH
Bremen, Germany
As electromagnetic radiation is severely attenuated in seawater, a submerged submarine almost entirely relies on acoustic sensors for long-range detection and surveillance. Because stealth is a submarine’s major advantage over other naval platforms, active operating sonars are rarely used by submariners. Instead, passive sonar arrays are used for long-range detection and ranging of other platforms.
A typical Atlas Elektronik Integrated Sensor Underwater System 90 for a modern conventional U 214-class submarine combines several antennae. The cylindrical hydrophone array (CHA), located in the bow of the submarine, is the main acoustic antenna array for the panoramic detection and tracking of targets in the medium-frequency domain. The flank array (FA), located on both sides of the submarine, is used for sound evaluation in the lower frequency band. The FA enables medium to long-range detection and classification of targets. The towed array (TA) is a flexible line array that can be deployed from a winch system in the rear part of the submarine. It is designed to achieve long-range detection with high bearing accuracy and excellent target separation at low frequencies.
The passive ranging sonar (PRS) is an antenna system and signal-processing method. With three antenna arrays fitted in a discontinuous line on the upper half of the submarine’s pressure hull, the incident acoustic wave front curvature is measured by a correlation process. With knowledge of the wave front curvature, the range to the target can be calculated. The performance of the PRS depends on the acoustic transmission conditions of the sea area of operation and on the target range. Surface ships and other naval platforms radiate broadband and narrowband noise from their propulsion systems and other machinery into the water column.
For the CHA, FA and TA, there are parallel broadband and narrowband signal processing chains. While the narrowband processing is essential for target analysis and classification, broadband detection gives an overview of the targets surrounding the ship.
Passive sonar provides bearing information that is put into the target motion analysis (TMA) module, which estimates the course, speed and range of the target under consideration. In order to derive target range estimates, bearing tracks of the detected targets of interest are formed. In this way, a tactical picture of the sea area of operation can be compiled. High-quality bearing tracks for the TMA algorithm are essential in order to minimize errors in the TMA solution. In current operational sonar systems, the initialization, maintenance and deletion of target bearing tracks is typically performed manually by the operator.
In this article, a multitarget tracking (MTT) algorithm for broadband passive sonar based on a multihypothesis tracking approach is presented. It is designed to automatically track all broadband sonar targets, thereby releasing the sonar operator from the tasks of track initialization, maintenance and deletion. The goal is to track all detectable targets, including weak, low signal-to-noise ratio targets, while at the same time minimizing the number of false tracks.
The following sections describe the MTT system and its algorithms, and some broadband-passive-sonar-specific features, important for an automatic tracking routine, are reviewed. Finally, the performance of the algorithm will be analyzed with the aid of simulations.
Multitarget Tracking Systems
Modern systems can be divided into single-target tracking (STT) and MTT systems depending on the number of simultaneously tracked physical objects. An integral part of any such system is a physical sensor (e.g., a radar or a sonar sensor). In the case of an STT, the sensor has a small field of view (FOV), and it follows the moving object during tracking.
In an MTT system, a large FOV is scanned by the sensor, and all objects in this region are processed at the same time. The sensor signal processing periodically generates lists of detections (e.g., detected radar echoes in the case of the active radar), which are passed on to the tracking unit. The task of the tracking unit is the logical combination of recurring sensor detections that emerged from one particular object and the estimation of its kinematic parameters. The fact that the list of scan detections not only contains measurements from objects of interest but also from false alarms creates a sophisticated challenge.
In many cases, the tracking unit is composed of three interrelated main elements. The state-estimation block is responsible for estimating the true path of an object by adding together the correct sequence of measurements. This is achieved in close relation with the data-association block, which deals with the correct assignment of newly available detections to already existing sequences of corresponding measurements. Included in the estimator is a motion model for the objects under consideration and a measurement model, which describes how the detections are related to the state of the objects whose parameters are to be estimated. For all target tracks, the motion model predicts where a new measurement should appear at the next sensor scan. The detections from the actual list are compared with the predicted measurements, and the data association routine creates a solution that assigns all detections from the list of measurements to tracks. All of the estimated paths formed by the interplay of these two building blocks are called tentative tracks, because the tracking system will treat them as candidates for real tracks.
In the management block, it is decided if a tentative track is to be extracted (confirmed), terminated (deleted) or remain tentative by applying a sequential probability ratio test, where the probability of the measurement sequence being in accordance with the motion model is set in relation to the probability that the respective sequence represents only false alarms.
Multihypothesis Approach
Under realistic conditions, the tracking system has to deal with a number of uncertainties, such as erroneous measurements, limited sensor resolution, imprecise knowledge of the object motion and high false alarm rates. For that reason, a correct assignment of new detections to existing tracks is not often possible in real-world applications, and it is more appropriate to use the multihypothesis approach for data association. In this approach, multiple new detections are assigned to a particular track and a hypothesis tree of possible measurements is built up. The idea is to simultaneously process more than one data interpretation history. To avoid the combinatorial disaster involved in applying this method (i.e., the number of hypotheses grows exponentially with the number of detections), the multihypothesis tracker (MHT) employs gating and pruning techniques that reduce the number of possible hypotheses.
Detection Algorithm
The main antennae used for panoramic detection and tracking are the CHA and the FA. From the viewpoint of the tracking system, these sensors are real MTT sensors with a large FOV. The quantity of interest processed in passive sonar is the incoming acoustic energy from all horizontal directions integrated over a certain interval of time and a certain (broad) frequency band. A beamforming operation is performed on the arriving signals in order to generate a discrete mapping of the surrounding sound intensity distribution. With knowledge of a submarine’s course, this can be transformed into a north stabilized energy distribution.
A dedicated detection algorithm then locates potential target bearings (bearing angles) by analyzing the given energy distribution. From this, a list of target detections for the respective time interval is created. This list contains true target detections as well as false alarms. Depending on the respective sonar array geometry and its location on board the submarine, a specific array exhibits one or more blind detection sectors. These are azimuthal directions from which either no detections occur or where the detections are severely erroneous. The CHA, for example, is obscured by the submarine, and therefore has a blind sector at the aft.
Testing the Algorithm
The performance of the MTT algorithm can be analyzed with simulations. A simulated scenario can be defined that contains several different surface vessels traveling along specified trajectories and a submarine’s track, typically including TMA maneuvers. Based on the trajectories of the target vessels and the submarine, a true-north-stabilized bearing time record (BTR) can be computed, giving the true bearing histories that a perfect bearing sensor on the submarine would produce. In this way, a number of crossings between bearing tracks belonging to different target vessels can be analyzed.
A simulation based on the true bearing histories takes into account the effects of the underwater sound channel and the imperfections of the acoustic sensor. The bearing detections extracted by the detection algorithm for each time-step make up the simulated BTR.
In the simulation, several typical features of broadband passive sonar appear: Among the detections originating from the true target vessels, there are a number of false detections. Bearing accuracy depends on signal-to-noise ratio and platform-relative bearing sector. Problems often arise during boat maneuvers when the blind sector of the sensor crosses the bearing track of a target. This can result in a distortion of the bearing tracks.
Another salient feature of passive sonar is that weaker targets are typically obscured by stronger ones due to the limited target separation capability of the sensor and therefore cannot be detected during target-crossing situations.
The list of detections for each time-step is then fed into the MHT tracker, which performs the estimation and data association steps. If a collection of recurring detections passes the sequential probability ratio test, the tentative track becomes extracted after a particular amount of time. Then a track number is assigned to the estimated track trajectory. A track becomes deleted if the probability of the underlying detections originating from a true target is low or if no more measurements could be found to extend the respective track.
The MHT usually extracts only tracks whose detections originate from a vessel present in the simulated scenario. Due to the imperfections of the sensor described above, the trajectory for a particular target is usually built up by several extracted tracks. The combination of the correct track parts, those originating from one particular target, can be handled manually by the sonar operator or can be computed automatically by applying a track-to-track assignment algorithm.
Conclusions
Using a multihypothesis tracking approach, the MTT algorithm presented in this article can help reduce the time operators of broadband passive sonar have to spend manually initializing, maintaining and deleting target bearing tracks.
As has been shown in the simulation testing described, the algorithm has the capability to automatically track all relevant targets without manual input by the user, and it therefore has great potential as a tracking system.
Kevin Brinkmann has been a senior systems engineer in the submarine systems division of Atlas Elektronik GmbH since 2007. His main focus is on multitarget tracking approaches in sonar applications. In 2006, he received his Ph.D. in physics from Göttingen University.
Jörg Hurka is a senior systems engineer in the submarine systems division of Atlas Elektronik GmbH, where he is responsible for submarine systems research and development. He received a Ph.D. in physics from the Rheinisch-Westfälische Technische Hochschule Aachen in 2002.
Sea Technology is read worldwide in more than 110 countries by management, engineers, scientists and technical personnel working in industry, government and educational research institutions. Readers are involved with oceanographic research, fisheries management, offshore oil and gas exploration and production, undersea defense including antisubmarine warfare, ocean mining and commercial diving.