This work investigates the behavior of a distributed team of agents on a dynamic distributed task allocation problem. Previous work finds that a distributed decision making process can effectively assign tasks appropriately to team members even when agents have only local information. We study this problem in a distributed environment in which agents can move, thus causing local neighborhoods to change over time. Results indicate that a higher level of adaptation is clearly required in the dynamic environment. Despite the increased difficulty, the distributed team is able achieve comparable behavior in both static and dynamic environments.

We examine the use of local decentralized decision-making methods for solving the problem of resource allocation. Specifically, we study the problem of frequency coverage given a team of cooperating receivers. The decision making process is decentralized in that receivers can only communicate locally, We use an extension of the minority game approach to allocate receivers to current frequency coverage tasks.

Finding certain associated signals in the modem electromagnetic environment can prove a difficult task due to signal characteristics and associated platform tactics as well as the systems used to find these signals. One approach to finding such signal sets is to employ multiple small unmanned aerial systems (UASs) equipped with RF sensors in a team to search an area. The search environment may be partially known, but with a significant level of uncertainty as to the locations and emissions behavior of the individual signals and their associated platforms. The team is likely to benefit from a combination of using uncertain a priori information for planning and online search algorithms for dynamic tasking of the team. Two search algorithms are examined for effectiveness: Archimedean spirals, in which the UASs comprising the team do not respond to the environment, and artificial potential fields, in which they use environmental perception and interactions to dynamically guide the search. A multi-objective genetic algorithm (MOGA) is used to explore the desirable characteristics of search algorithms for this problem using two performance objectives. The results indicate that the MOGA can successfully use uncertain a priori information to set the parameters of the search algorithms. Also, we find that artificial potential fields may result in good performance, but that each of the fields has a different contribution that may be appropriate only in certain states.

This work investigates the efforts behind defining a classification system for multi-agent search and tracking problems, specifically those based on relatively small numbers of agents. The pack behavior search and tracking classification (PBSTC) we define as mappings to animal pack behaviors that regularly perform activities similar to search and tracking problems, categorizing small multi-agent problems based on these activities. From this, we use evolutionary computation to evolve goal priorities for a team of cooperating agents. Our goal priorities are trained to generate candidate parameter solutions for a search and tracking problem in an emitter/sensor scenario. We identify and isolate several classifiers from the evolved solutions and how they reflect on the agent control systems's ability in the simulation to solve a task subset of the search and tracking problem. We also isolate the types of goal vector parameters that contribute to these classified behaviors, and categorize the limitations from those parameters in these scenarios.

In this paper, we propose a rule-based search method for multiple mobile distributed agents to cooperatively search an area for mobile target detection. The collective goals of the agents are (1) to maximize the coverage of a search area without explicit coordination among the members of the group, (2) to achieve sufficient minimum coverage of a search area in as little time as possible, and (3) to decrease the predictability of the search pattern of each agent. We assume that the search space contains multiple mobile targets and each agent is equipped with a non-gimbaled visual sensor and a range-limited radio frequency sensor. We envision the proposed search method to be applicable to cooperative mobile robots, Unmanned Aerial Vehicles (UAVs), and Unmanned Underwater Vehicles (UUVs). The search rules used by each agent characterize a decentralized search algorithm where the mobility decision of an agent at each time increment is independently made as a function of the direction of the previous motion of the agent, the known locations of other agents, the distance of the agent from the boundaries of the search area, and the agent's knowledge of the area already covered by the group. Weights and parameters of the proposed decentralized search algorithm are tuned to particular scenarios and goals using a genetic algorithm. We demonstrate the effectiveness of the proposed search method in multiple scenarios with varying numbers of agents. Furthermore, we use the results of the tuning processes for different scenarios to draw conclusions on the role each weight and parameter plays during the execution of a mission.

This paper will compare the various methods of analyzing the results of radar pulse train deinterleavers. This paper is divided into three sections. The first section of this paper will describe the basic methods, such as the confusion matrix, and some measures that can be obtained from the matrix. The measures will include correct correlation, miscorrelation, ambiguity and track purity. Correct correlation is calculated by dividing the total number of correctly clustered pulses by the total number of pulses in the collect. Miscorrelation measures the fraction of received pulses that incorrectly deinterleaved. Ambiguity measures the fraction of received pulses that are rejected by the deinterleaver as having uncertain association with a ground truth track. Track purity measures the ability of the deinterleaver to create a constructed track comprised of pulses from a single ground truth track. These metrics will show the quality of the deinterleaving operation. The second section of this paper will describe some of the advanced similarity measures of effectiveness. This section will also describe how distance measures will be used to analyze deinterleaver results. The two main similarity measures to be discussed in this paper will be the Rand Adjust and Jaccard coefficient. These similarity measures are also known as criterion indices and are used for evaluating the capacity to recover true cluster structure. The reason for the selection of the Jaccard and Rand Adjust as measures is that they both allow a value to be obtained that is between 0 and 1 that will show how good the clusterer in question has performed. The Rand Adjust also allows for more variability in the range between 0 and 1 and appears to provide a more accurate evaluation of the cluster. The distance measures that will be described include Euclidean, Mahalanobis and Minkowski distances. These distance measures have different methods to evaluate each cluster for purity. These measures will provide an indication of the quality of the deinterleaver operation. The third section of this paper will provide the results from these measures on a deinterleavered dataset and will discuss the comparison of these metrics.

We present a genetic algorithm (GA) based decision tool for the design and configuration of teams of unmanned ground sensors. The goal of the algorithm is to generate candidate solutions that meet cost and performance constraints. The GA evolves the membership, placement, and characteristics of a team of cooperating sensors. Previous work shows that this algorithm can generate successful teams in simple, obstacle free environments. This work examines the performance of our algorithm in environments that include obstacles

This paper provides a comparison of the two main techniques currently in use to solve the problem of radar pulse train deinterleaving. Pulse train deinterleaving separates radar pulse trains into the tracks or bins associated with the detected emitters. The two techniques are simple time of arrival (TOA) histogramming and multi-parametric analysis. TOA analysis uses only the time of arrival (TOA) parameter of each pulse to deinterleave radar pulse trains. Such algorithms include Cumulative difference (CDIF) histogramming and Sequential difference (SDIF) histogramming. Multi-parametric analysis utilizes any combination of the following parameters: TOA, radio frequency (RF), pulse width (PW), and angle of arrival (AOA). These techniques use a variety of algorithms, such as Fuzzy Adaptive Resonance Theory (Fuzzy-ART), Fuzzy Min-Max Clustering (FMMC), Integrated Adaptive Fuzzy Clustering (IAFC) and Fuzzy Adaptive Resonance Theory Map (Fuzzy-ARTMAP) to compare the pulses to determine if they are from the same emitter. Good deinterleaving is critical since inaccurate deinterleaving can lead to misidentification of emitters.The deinterleaving techniques evaluated in this paper are a sizeable and representative sample of both US and international efforts developed in the UK, Canada, Australia and Yugoslavia. Mardia [1989] and Milojevic and Popovich [1992] shows some of the early work in TOA-based deinterleaving. Ray [1997] demonstrates some of the more recent work in this area. Multi-parametric techniques are exemplified by Granger, et al [1998] and Thompson and Sciortino [2004]. This paper will provide an analysis of the algorithms and discuss the results obtained from the referenced articles. The algorithms will be evaluated for usefulness in deinterleaving pulse trains from agile radars.

Purity of deinterleaved pulse descriptor word (PDW) trains is critical to the performance of emitter classification software that analyzes PDW data. Contamination of the input PDW train can lead to artifacts in the analysis resulting in incorrect or ambiguous classification. This paper presents results of an investigation into the possibility of applying transition matrices to the detection and removal of contaminating pulses from a deinterleaver output. The utility of transition matrices as a fast and efficient pattern recognition tool in ESM emitter classification has been known for over a decade [Brown, R,G., "Pattern Recognition Using Transition Matrices", 7(th) RMC Symposium on Applications of Advanced Software in Engineering, pp 52-57, Royal Military College of Canada, Kingston Ontario, May 1995]. In the work presented herein, transition matrix patterns unique to contaminated pulse trains are sought in order to provide a warning that a particular PDW train is contaminated, and provide a clue as to which pulses should be removed to purify the PDW train.

Improvements in sensor capabilities have driven the need for automated sensor allocation and management systems. Such systems provide a penalty-free test environment and valuable input to human operators by offering candidate solutions. These abilities lead, in turn, to savings in manpower and time. Determining an optimal team of cooperating sensors for military operations is a challenging task. There is a tradeoff between the desire to decrease the cost and the need to increase the sensing capabilities of a sensor suite. This work focuses on unattended ground sensor networks consisting of teams of small, inexpensive sensors. Given a possible configuration of enemy radar, our goal isto generate sensor suites that monitor as many enemy radar as possible while minimizing cost. In previous work, we have shown that genetic algorithms (GAs) can be used to evolve successful teams of sensors for this problem. This work extends our previous work in two ways: we use an improved simulator containing a more accurate model of radar and sensor capabilities for out fitness evaluations and we introduce two new genetic operators, insertion and deletion, that are expected to improve the GA's fine tuning abilities. Empirical results show that our GA approach produces near optimal results under a variety of enemy radar configurations using sensors with varying capabilities. Detection percentage remains stable regardless of changes in the enemy radar placements.

High-accuracy, low-ambiguity emitter classification based on ESM signals is critical to the safety and effectiveness of military platforms. Many previous ESM classification techniques involved comparison of either the average observed value or the observed limits of ESM parameters with the expected limits contained in an emitter library. Signal parameters considered typically include radio frequency (RF), pulse repetition interval (PRI), and pulse width (PW). These simple library comparison techniques generally yield ambiguous results because of the high density of emitters in key regions of the parameter space (X-band). This problem is likely to be exacerbated as military platforms are more frequently called upon to conduct operations in littoral waters, where high densities of airborne, sea borne, and land based emitters greatly increase signal clutter. A key deficiency of the simple techniques is that by focusing only on parameter averages or limits, they fail to take advantage of much information contained in the observed signals. In this paper we describe a Dempster-Shafer technique that exploits a set of hierarchical parameter trees to provide a detailed description of signal behavior. This technique provides a significant reduction in ambiguity particularly for agile emitters whose signals provide much information for the algorithm to utilize.

This paper will evaluate one promising method used to solve one of the main problems in electronic warfare. This problem is the identification of radar signals in a tactical environment. The identification process requires two steps: clustering of collected radar pulse descriptor words and the classification of clustered results. The method described here, Fuzzy Adaptive Resonance Theory Map (Fuzzy ARTMAP) is a self-organizing neural network algorithm. The benefits of this algorithm are that the training process is very stable and fast and that it needs a small number of required initial parameters and it performs very well at novelty detection, which is the classification of unknown radar emitters. This paper will discuss the theory behind the Fuzzy ARTMAP, as well as results of the processing of two `i real^i radar pulse data sets. The first evaluated data set consists of 5242 radar pulse descriptor words from 32 different emitters. The second data set consists of 107850 pulse descriptor words from 112 different emitters. The radar pulse descriptors words that were used by the algorithm for both sets of data were radio frequency (RF) and pulse width (PW). The results of the processing of both of these datasets were better than 90% correct correlation with actual ID, which exceeds the results of processing these datasets with other algorithms such as K-Means and other self-organizing neural networks.

U) (U) This paper will describe the application of improved methods of sorting, classifying, and identifying pulsed and modulated CW command and control (C2) emitters, whether air, land, sea, or subsurface. These methods are based on Artificial Intelligence (AI) and include: • Probabilistic reasoning techniques such as Bayesian and/or Dempster-Shafer theory for data fusion and analysis. • Genetic algorithms for multi-dimensional search and optimization (i.e. to search for optimized maneuvering and threat-response tactics.) • Fuzzy set theory, which can be implemented with various other AI techniques. For example, fuzzy logic controllers may be coupled to genetic algorithms and knowledge based rule-generating systems to provide adaptive control mechanisms. • Neural networks, to aid in solving many pattern recognition, decision-making, data fusion and analysis problems. (U) We will then discuss the integration of the results from these techniques with software agents, to support the capability to continuously manage the influx of on-board sensor data with received off-board information to maintain situational awareness (the relationship between ownplatform and all other observed platforms in the context of the common physical environment). Software agents are characterized as autonomous, intelligent, collaborative, and perhaps exhibit some degree of mobility, acting on a user’s behalf in order to meet specific objectives. Lastly, we will define the concept of a mission management system that uses the developed situational awareness to perform targeting, avoidance, and denial (countermeasures) functions of ownplatform against other-platforms, while assessing own-platform states such as weapon loads and electromagnetic combat capability. A further function of the mission management system would be to specify ingress/egress routing.

In the Network-Centric Warfare (NCW) paradigm, battlespace agents autonomously perform selected tasks delegated by actors/shooters and decision-makers including controlling sensors. Network-Centric electronic warfare is the form of electronic combat used in NCW. Focus is placed on a network of interconnected, adapting systems that are capable of making choices about how to survive and achieve their design goals in a dynamic environment. The battlespace entities: agents, actor/shooters, sensors, and decision-makers are tied together through the information and sensors grids.

Searching a large visual database for the item(s) most similar to a test object is a tedious task that has many applications. This task is further complicated if the test object (extracted from an image) is degraded by its distance and other factors. A distant target is typically subject to blurring, noise, and other distortions, so that when it is matched to database items it: may turn out to be less similar to its own ideal model than another database item. To reduce the likelihood of such misidentifications, we propose an approach where the database is enlarged to include appropriately degraded versions of the ideal items in the original database. The extracted target is likely to match closely one of the distorted versions of its corresponding database item. In particular we discuss the problem of ship identification from the smokestack insignia, i.e.. extracting the ship's insignia from its color photo and finding its closest matches in a database containing thousands of registered shipping line insignias. This is a problem of interest to international regulatory agencies, insurance companies and others. We discuss how the database insignias, as well. as the extracted insignias, are processed and indexed, and how the distorted versions of ideal insignias are generated.

ABSTRACT Electronic Warfare (EW) Support Measures (ESM) systems are evolving rapidly toward unmanned status by using a constellation of artificial intelligence (AI) technologies. Among these are artificial neural networks (ANN), genetic algorithms, intelligent databases, fuzzy logic and data fusion techniques, as well as expert systems. The applications and implementation techniques, for which adaptive autonomous non cooperative target recognition is useful or necessary, form the focus of this discussion. Military and civilian applications abound for fixed site, shipborne, and airborne systems. These systems could control and manage other interactive systems normally requiring manual inputs. Applications such as air and marine traffic control and management regimes, collection of transport traffic statistics for economic analysis and intelligence, passive targeting, platform inventory (order of battle), and anti‐fratricide control are examples meeting this requirement.ESM systems use these AI technologies throughout the signal processing chain and beyond: from acquisition through automatic RF band scheduling and antenna pointing, as well as deinter‐leaving of dense signal environments and characterization of individual emitters, to rapid, high‐confidence emitter‐platform correlation with automatic optical/IR imaging, and cross‐indexing of multiple worldwide information databases to prevent platform aliasing. These systems are now feasible because the required components are now available. Low‐cost miniaturized RF system and subsystem components, massively parallel computing resources, and ANN integrated circuits permit the achievement of low‐cost, low‐power, small, modular systems that stand and act alone.

This paper presents a method of constructing a low dimensional representation of a pulse signal that preserves the information necessary to classify the pulse. A large set of "atomic features" are generated from dilations and shiftings of a family of one or more mother wavelet functions. Each atomic feature is used to generate a feature vector element by taking the inner product of the pulse with the atomic feature. The "best" feature vector elements are selected according to the mutual information between the pulse category and feature vector element values over all pulse realizations. The result is a low dimensional transformation that retains the information necessary to discriminate one category from another. This paper presents an algorithm for computing the above mutual information measure on a CNAPS, a 512 processor SIMD parallel computer.