Integrated underwater surveillance system.

Arne Løvik, Arnt Rune Bakken, Endre Marken and Martin August Brinkmann Kongsberg Defence & Aerospace as

Kirkegårdsveien 3, 3600 Kongsberg, Norway

Abstract — The C-Scope integrated surveillance is a well established system in the vessel traffic management area and has over the past years been complemented with a novel underwater surveillance and protection part with a layered system approach. The new development of both active sonar systems for longer range and passive systems for large area detection and classification has been completed. The paper introduces the new wide-band sonar units and gives some insight to the implemented classification concept. The active sonars cover the frequency range from 3-8 kHz and 30-40 kHz, while the passive system is from 10-5000 Hz. The paper will concentrate on the active part of the systems and gives examples with results from some installations.

Keywords-component; surveillance system, sonar

I. INTRODUCTION New changes to port security address not only the

normal or friendly traffic, but also systems that secure the normal port operation, and deter and prevent any hostile operation against the harbor.

The mere cost of an incident that halts normal operation of a major port will incur significant economical and operational consequences to a country or region. In this paper we will discuss threats from intruders, such as surface vessels, underwater vehicles, divers and swimmers, all being targets that will hide their real identity and intent.

Diver detection sonar systems with detection ranges from some hundred meters to 600-800 meters on a good day have been available for some time. But before countermeasures or reaction procedures can be set into operation, the security system will have to perform all the tasks from detection, classification, alerting the operator to confirmation of the threat being real by the operator before the operator chooses the best reaction procedure for the incident.

Studies have shown that the time to perform these tasks will be very limited, and thus work was undertaken to develop sonar and systems with larger detection ranges and integrate them into a system allowing the timely performance of port security.

The project discussed in this paper began in 2007, aiming to develop and test new long-range awareness sonar systems in a real environment at Haakonsvern Naval base, which could form the basis for a permanent installation at the base.

The sonar was to be integrated into a system including other above- and below-water sensors and reactors.

II. ABOVE WATER SURVEILLANCE Protection against illicit underwater activities in

coastal areas, ports, harbors or in confined areas is complicated by the reverberant conditions, and the normally high levels of surface traffic.

To ease the burden on the operator, the system should only give a warning when something is unusual or abnormal and requires attention. This puts severe requirements on the signal processing and the ability to reliably reduce the number of 'false alarms' (i.e. events not requiring the attention of the operator.)

It is therefore very important to combine above-water information with sonar information in order to improve on classification and thus reduce false alarms. There are many VTS (Vessel Traffic Service) and Port Management Systems installed in important ports around the world. These systems already handle all important above-water sensors. Therefore, it was decided to integrate the sonar information in an existing platform, the C-Scope Vessel Traffic Management System, in order to provide a common operational picture on track level, for targets above and below the surface.

The system operates 24/7 with a minimum number of operators. It was therefore vital that all information from above- and underwater sensors are collected, sorted, tracked and fused.

Fused tracks are compared with ship data in databases, and alarms of underwater activities are given only for sonar tracks that pass alarm zones and do not correlate with expected targets on the surface. Based on the results form the trials the following features are recommended:

• Open system architecture.

• GIS (Geographic information system) that shows tracks and detections in geo-referenced standard sea charts and land maps in both 2D and 3D.

• Map layers that can be used by the operator to display available information for the area.

• Advanced sensor fusion covering all above- and underwater sensors.

Fig. 1 shows an example of fused tracks, giving the operator a coherent view of all large surface vessels in the area.

Figure 1. C-Scope Tracking service

It is also important to be able to study an incident in detail, for example to investigate the cause of an automatic alarm. The C-Scope therefore has Recording and Replay service taps that feed system data back exactly as it appeared, with complete detail of the following:

Radar, sonar video Radar, sonar tracks,AIS tracks and fused tracks Operator inputs Images from the CCTV system Audio from the maritime communications system Bearing lines from the direction finders.

III. UNDERWATER SURVEILLANCE AND PROTECTION (USP)

An important factor in underwater surveillance and protection is providing sufficient time for the operator to react.

To achieve maximum time for reaction a layered approach is often considered. An example is shown in the Fig. 2.

Figure 2. Layered approach with three zones.

The threats perceived in the different zones are normally

Inner zone: divers, UUV, scooters, RIB Middle zone: DDV, UUV, Midgets, RIB Outer zone: submarines. UUV, Midgets, other The targets to be detected are generally smaller and

slower as we move towards the final assets in the inner zone.

In addition to the zone structure, two other approaches are often found, one is the total volume coverage and the second being control of choke points.

For the early warning and preparedness, the more layers the more awareness is obtained.

The choice of the volume or choke approach is often one of cost and severeness of the threat to the area in question.

For a system to be complete different sensors will be required for the different zones.

The other zone is more like ASW type sonar, while the inner zone by many has been compared to MCM operations.

Further different reaction measures must be included to move from a surveillance system to a protection system. In this paper the reaction side of the system will not discussed, however the KONGSBERG C-Scope system may be fitted with reaction units from the soft warning to the hard kill type.

Going back to the sensor side the range of application from ASW to MCM creates a set of different sonars dedicated for operation in the near coastal environment and with the requirement of having extremely low false alarm rate and with a high degree of automation in the decision or classification process.

For the coastal and harbor surveillance KONGSBERG has developed in cooperation with FFI and the RNoN a set of new wide-band sonar systems.

The active sonar is called LASAR, Long Range Awareness sonar and the passive counterpart PASAR.

The PASAR operates in the frequency range 10-5000 Hz while LASAR has different frequency bands, 3-8kHz, 10-20kHz and 30-45kHz.

The active sonar LASAR can operate in active and passive mode simultaneously, or separately. For the LASAR 40, active band is 30-45kHz and the passive band covers 1-45kHz. In combined mode the passive part is mainly used for detection of fast going smaller surface crafts and for classification on its own or in combinations with the active retrieved information.

From this discussion it should be clear that the LASAR 5, 3-8kHz sonar is aimed for the outer zones giving long range detection of larger targets such as submarines, UUV and midgets. The PASAR is often used in combination with the LASAR 5. The transducer array of the LASAR 5 is a linear array giving angular resolution of 0,7 degrees and less.

In the following a discussion of the LASAR system is given exemplified by the LASAR 40.

IV. LONG RANGE AWARENESS SONAR: LASAR 40

The LASAR 40 was developed in the project as a part of the test installation at Haakonsvern Naval base in Bergen, Norway. The main goal was to detect small targets such as divers or unmanned underwater vehicles (UUVs) consistently, at a range of 1000 meters or more

The sonar is designed to be installed at fixed points in a port, or alternatively as a moveable containerized system for port security, or as a portable device for fleet protection. To help in classifying targets, the system includes a passive chain in parallel to the active part.

Internal studies [2] and others [1] show that the 'optimal' frequencies for this type of application are more in the range of 30 to 40 kHz than around 80 to 100 kHz, which is today commonly used for diver detection sonar.

Another variable in a busy port is the wake from passing vessels.

Many studies have been performed on the effects of bubbles in the water and the caused attenuation. Some of the effects are related to the bubbles generated by the wave motion, i.e. related to the sea state. The general conclusion is clear: the higher the frequency, the more the effect on sound propagation. This follows from the fact that the smaller bubbles live longer and are more abundant than the larger ones.

To preserve the portability and adaptability to different geometries, a modular system was conceived with a single separate transmitter and a number of linear receiving arrays. The receiver may be configured to provide a 60 to 360 degree field of view, without any loss or difference in the performance. The flexibility in the number of receiving arrays makes it possible to optimize the sonar to environmental conditions in different ports.

However the configuration with linear array forming the receiving array for the different geometries imposes

new and challenging problems for the design of the beamformer and a novel and flexible approach has been developed [3]. The adopted projection method with suitable tapering works extremely well with these nonlinear array.

Longer arrays may be used in ports with very high reverberation, and the bandwidth can be chosen depending on the environment, giving a range resolution from a few cm and up. The sonar is therefore very adaptive to different environmental conditions.

The frequency bands covered are:

Active: 30 - 45 kHz

Passive: 1- 30 kHz

The transducer unit may be mounted on the seabed, over the side of a vessel or as a pier mount. The processing unit is made in one compact portable unit suited to meet the operational requirements. The first two units were installed at Haakonsvern Naval base and the next section presents some of obtained test results.

V. CLASSIFICATION The over all classification scheme consists of two

separate branches:

• one related to attributes associated to the target as such

• the other to the behaviour of the target specially in relation to the perceived threat.

The target related attributes are typically among the following

• echo strength

• echo variation with aspect

• echo variation with frequency

• target speed

• target acceleration

• target turn rate

• emitted noise

• noise spectrum or characteristics

To define the attributes above from the returned echoes is in itself an interesting task, the target strength for instance could be related to the peak value in and an area of returned energy or to the average of some sort. Further a number of additional parameter could be defined related to the echo mass; its extent, major axis, perimeter or others.

The various attributes will have at least two values, a static on and a temporal. The static giving the value for the attributes per ping and the temporal giving some measure of it temporal variation or statistical behavior.

For the classification process the most important factor is to select those attributes that will give the “best” classification result for the group of targets involved. Best classification will normally mean a classification that will yield high rate of correct classification and low rate of false classification.

Figure 3. Scatter plots with density estimation, (a) for a static set and (b) for a temporal set for object classes diver and marine life.

5

Two example of a set of such attributes are shown in the Fig. 3. The Fig. 3 (a) for a static set and (b) for a

temporal set.

The values of the attributes related to the target can be used in a probability based classifier such as Naïve Bayes classifier. Bayesian classifiers assign the most probable class to a given example described by its attribute vector. Learning such classifiers can be greatly simplified by assuming that the attributes are independent variables for the given class.

Let X = (x1, ..., xn) be a vector of observed random variables, called attributes. The Naïve Bayes classifier will then give the most probable class Ci. The probability of having class Ci given the attributes X, is by Bayes’ theorem;

.

The conditional probability P(X|Ci) is the predictions that the model makes about the data X when its parameters have a particular value x. The “prior” distribution P(Ci) states what values the model’s parameters might plausibly take. The normalising factor P(X) is only based on the values the attributes take on.

Bayesian Classifiers are known to be the optimal classifiers, since they minimize the risk of misclassification. However, they require defining P(X|Ci), i.e. the joint probability of the attributes given the class. Estimating this probability distribution from a training dataset is a difficult task, since it may require a very large dataset even for a moderate number of features in order to significantly explore all the possible combinations.

In the framework of the Naïve Bayes Classifier, the attributes are assumed to be independent from each other given class. This allow us to write the P(X|Ci) as:

.

The Naïve Bayes Classifier is therefore fully defined by the conditional probabilities of each attribute given the class.

When it comes to the behavior pattern it will normally be related to the threat the behavior imposes. Typical patterns for a diver with a dedication to reach a goal, an erratic behavior of a school of fish, the tidal variation of clutter and many more.

A first simple but efficient measure of the behavior is the travel distance compared to the distance travelled towards the area of protection, high value asset. This measure K can be defined by

,

where v is the speed of the target and vn is the component of the velocity vector towards the high value asset, positive towards the asset, negative away.

All negative values of K is not threatening behavior, likewise zero is no problem. When the K value approaches 1 the behavior is clearly hostile. Fig. 4 shows experimental values of the K for unknown targets and for diver threats. The figures show that K for diver threats has a narrow distribution around 1, while K has a much wider distribution for unknown targets.

VI. SONAR AND SYSTEM TESTS The installation and test at the Norwegian

Haakonvern Naval base has now been underway since early 2007, with different test systems and sonar heads. The latest is the prototype LASAR 40 of which two heads are installed along the main pier. The transducers are bottom mounted.

Tests have been devised to look at detection range against various targets, their aspect and the seasonal variations, among others. The achieved detection ranges have been compared with predictions using the sonar performance model LYBIN, supplied by the Royal Norwegian Navy [4].

In general, there is a good correspondence between the simulations and the obtained results. There is a consistent detection to 1100 meters or more against divers with closed or semi-closed systems with the prototype of the LASAR 40.

The production model is now under delivery to the YUNUS project in Turkey for the protection of their two largest naval bases. This integrated underwater surveillance system is based on a volume coverage approach and counts many sonars to cover the area in mono and bi-static mode of operation.

Fig. 5 shows an illustration from another installation where both diver and mammals are detected and tracked correctly.

Figure 4. K factor for unknown targets and for diver threats.

6

Figure 5. Operator display with diver (red) and dolphins (green).

VII. SUMMARY This paper has presented a set of new sonar systems,

active and passive for coastal and port surveillance. The LASAR 40 sonar is lower in frequency than standard diver detection sonar and gives longer detection ranges than those normally achieved. The major benefit of longer range come in the volume coverage approach to surveillance as opposed to a choke point approach where range necessarily is a prime factor.

The low frequency LASAR 5 gives large volume coverage in coastal areas for detection of larges targets, submarines to midgets.

To reduce the burden on the operator a sophisticated classifier has been introduced. The classifier uses attributes associated with the target itself and with the behavior of the target.

Tests of the systems in different waters and environmental conditions show very robust and good detection and classification performance for the automatic sonar processing system.

REFERENCES [1] M. G. E. D. Colin, S. P. Berens and M. A. Ainsie, (TNO),

“Optimal frequency for diver detection sonar,”. [2] A. Løvik, H. Aagedal, E. Mathisen and E. Marken (Kongsberg

Defence & Aerospace). ”Examples of the Combined Use of Active and Passive Sonar for Underwater Harbour Surveillance,” UDT UK, 2008.

[3] R. Otnes (FFI) and Nanna Skjei (Kongsberg Defence & Aerospace), “Projection-based tapering for conventional beamforming on nonlinear sonar arrays,” Oceans USA, 2010.

[4] K. T. Hjelmervik (FFI), E. M. Dombestein (FFI), T. S. Såstad (FFI), J. Wegge (FFI) and S. Mjølsnes (NDLO), “The Acoustic Raytrace Model Lybin – Description and Applications,” UDT UK, 2008.