© EYEWIRE Knowledge-Based Resource Management for … · 2015. 7. 19. · ness in radar...

In recent years, phased array antenna technology has beenmaturing rapidly, and this form of transduction is set tobecome the norm in complex and advanced radar systems.Instantaneous, adaptive beam pointing enables the combi-nation of functions such as tracking, surveillance, and

weapons guidance, which previously were performed by sepa-rate, dedicated individual radar systems. However, while beingable to instantaneously and adaptively position and control thebeam has clear advantages, it also brings about a new set ofchallenges. In particular, this form of radar is able to adapt itsparameters according to the way it perceives its environment.Thus, the interpretation of radar backscatter and the subsequent

decisions as to what and how radar resources are to be allocatedbecomes vital to unleashing the full potential of a phased arraysystem. This dynamic and interactive interplay between the set-ting of radar parameters to optimize the tasks to be carried outand perception of environment highlights the role in whichknowledge and intelligence will be central in multifunctionradar performance. For example, the system has to share itstime between all the different types of functions that it has toperform. This means that in certain operating conditions, it maynot have enough resources to perform all required tasks. In thisinstance, a control system has to evaluate the allocation ofresources to provide the best solution. The radar resource

IEEE SIGNAL PROCESSING MAGAZINE [66] JANUARY 2006 1053-5888/06/$20.00©2006IEEE

[Sergio Miranda, Chris Baker, Karl Woodbridge, and Hugh Griffiths]

Knowledge-BasedResource Managementfor Multifunction Radar

[A look at scheduling and task prioritization]

© EYEWIRE

Authorized licensed use limited to: University College London. Downloaded on October 6, 2008 at 10:11 from IEEE Xplore. Restrictions apply.

IEEE SIGNAL PROCESSING MAGAZINE [67] JANUARY 2006

management problem therefore reduces to the allocation offinite resources in an optimal and intelligent way. In this article,we consider two related aspects of radar resource management,scheduling and task prioritization. Two different methods ofscheduling are examined and compared and their differencesand similarities highlighted. The comparison suggests that pri-oritization of tasks plays a dominant role in determining per-formance. A prioritization scheme based on fuzzy logic issubsequently contrasted and compared with a hard logicapproach as a basis for task prioritization. The setting of prioritiesis shown to be critically dependent on prior expert knowledge.

The efficient allocation of radar resources such as time andenergy has been the subject of increasing research over recentyears. Since the problem is complicated and multidimensional,it is sensible to make use of whatever prior information is avail-able to form a knowledge-based solution. Most studies haveincluded different techniques such as artificial neural networks,decision theoretics, information theory, and mathematical pro-gramming techniques including linear, nonlinear, and dynamic.Radar resource management approaches have been divided intothree areas: adaptive track updates [1], search scans [2], andscheduling. Here, we concern ourselves with scheduling and itsclose relation task prioritization.

Scheduling is an important subproblem of radar resourcemanagement and has been the focus of intensive research inorder to optimize the process of accomplishing a set of measure-ments to be performed by a multifunction radar system, e.g.,[3]–[8]. These scheduling approaches are dissimilar in designand methodology, and little has been reported about the differ-ences in their performance in resource allocation when utilizedin similar operational conditions [9]. In addition, while mostapproaches have used fixed priority orders for ranking radartasks, few analyses have addressed the development and theeffects of adaptive methods for priori-tizing radar tasks as a function ofchanging tactical scenarios [10]–[12].Other notable papers that have exam-ined scheduling and closely relatedtopics include [16]–[21].

Closely coupled to scheduling andfundamental to successful systemperformance is task prioritization.Indeed, in any system where moretasks are demanded than there istime to execute, some form of priori-tization is mandatory. In real-timesystems, a prioritization is said to befeasible if the resulting system isschedulable. Thus, solving the priori-ty assignment problem means deter-mining a feasible priority assignmentfor a given system. Priority is usuallyassociated with the urgency andimportance of a task and a systemmust prepare a sequence able to exe-

cute ready tasks that have the highest priority. Most algorithmsare classified according to how priorities are assigned as a func-tion of time. Thus, the algorithms can be classified as fixed,dynamic, or of mixed priority. A fixed-priority algorithm evalu-ates all priorities at the design time, maintaining them for thelifetime of the task or mission. This is the simplest form of algo-rithm and is typically used for assigning priorities in schedul-ing algorithms. Conversely, a dynamic-priority algorithmassigns priorities online, based on execution parameters oftasks, such as upcoming deadlines. Finally, mixed-priority algo-rithms have both static and dynamic components. This articleinvestigates and presents methods based on fuzzy logic for pri-oritizing radar tasks with the aim of analyzing their effective-ness in radar scheduling.

SCHEDULINGTwo different scheduling algorithms proposed in [3] and [4] arefirst briefly described prior to comparing and contrasting theirperformance. First, we introduce a model for the phased arrayradar system to be examined. The model of the multifunctionradar developed here is focused on tracking, surveillance, andtask scheduling. The architecture used in the simulations is pre-sented in Figure 1.

A radar system composed of four fixed-phased array anten-nas is considered, and the simulation covers the behavior ofone of its faces. The boundaries of the volume of coverage ofthe face of the radar can be user defined. Based on this infor-mation and on the desired surveillance performance, the sur-veillance function calculates the number of radar beamsnecessary to survey that volume. A list of task requests is gen-erated, taking into account the desired surveillance perform-ance of the radar system. The surveillance manager is fed bythe task list, maintained an inactive queue of tasks (not yet

[FIG1] Radar resource manager architecture.

Operator andStrategy

WaveformDatabase

RadarFunctions

Model of Environment

Resource Manager

TX/RX

TrackManager

SurveillanceManager

PriorityAssignment

Scheduler


scheduled), and provides the scheduler with a smaller queue ofrequests that are close to their due time of execution.Likewise, the track function calculates the update times of thetargets under track and feeds the track manager with a queueof tasks to be scheduled. The track manager also maintains alist of inactive track tasks that are to be sent to the schedulerwhen close to their due time of execution. Both surveillanceand track manager select from the waveform database theparameters to be used in the transmission of the radar pulsesassociated with each radar job. Several aspects are to be takeninto consideration in the criteria of selection, includingboundaries of the regions of coverage, expected targets, targetrange, and target speed. Decisions relating to how resourcesare to be allocated according to the relative importance of thetasks and how to assess this relative importance are made bythe block called Priority Assignment. The scheduler is fed by

queues of track, plot confirmation, and surveillance taskrequests and creates a set of measurement tasks to be carriedout by the radar based on task priorities and time constraints.A feedback loop between the output of the scheduler and theradar functions enables the next update times related to thosetasks to be calculated. The overall preferences related to mis-sion requirements and resource management decisions basedon evaluation of the tactical scenario are accounted for in theblock named Operator and Strategy. This module operates as ahuman-machine interface, allowing intervention to enablecorrections in the behavior of the system.

Two different schedulers were developed and are comparedfor the same operational conditions. They are based on theonline scheduling algorithms presented by Orman et al. [4] andButler [3]. The scheduler suggested by Orman et al. is centeredon a coupled-task specification of the radar jobs. A coupled taskis a job consisting of two different operations separated in timeby a specified interval; for example, the transmission and recep-tion times of a defined waveform type. Thus, each coupled taskcan be represented by the processing time of the first operation,the processing time of the second operation, and the separationtime between them. The job is considered to be a set of tasksthat must be executed to achieve the radar function perform-ance requirements (e.g., a pulse burst). The scheduler organizesa queue of tasks to be executed in any order provided that twotasks do not occupy the radar at the same time. The main idea ofthe coupled-task scheduler is to use the idle time existing withina radar job to interleave other radar jobs and achieve improvedusage of the radar time.

An alternative approach was postulated by Stafford [14] andsubsequently modified by Butler [3] that is based on the conceptof time balance and was implemented in the experimental mul-tifunction electronic scanned array radar (MESAR) system.MESAR is an experimental active phased-array radar systemdeveloped by QinetiQ and AMS Ltd. The first version of thisradar was delivered in the early 1990s, and advances have beenintroduced more or less continuously since then.

The Butler (or MESAR) algorithm uses a time-balancescheme to control the scheduling process of the requested tasks.There is a time balance related to every radar job, indicating tothe scheduler how much time is owed by the radar to that job. Azero time balance meant that the job is due to be executed at theexact time. A negative time balance meant that the new job isnot to be executed at this time, and a positive time balancemeans that the job is late, thus the radar owes time to that job.A job consists of several tasks and is usually associated with sur-veillance of a region of coverage or the keeping of a target undertrack. A task is a group of activities that are noncoherently inte-grated to give detection. A task can be divided into looks, con-sisting of one or more activities. This algorithm schedules thelooks of a task in sequence and the tasks are selected accordingto the desired priority order. All radar functions are assumed tohave fluid deadlines. When a task of higher priority than thatcurrently being scheduled is requested, the process is interrupt-ed, and that task is scheduled. An interesting characteristic of


[FIG2] Results from the Butler type scheduling of surveillanceand tracking tasks in three sectors of coverage. Requiredsurveillance occupancy = 80%.

0 1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8

1

Time (s)

Occ

upan

cy

Region 1Region 2Region 3TrackingTotal Occupancy

[FIG3] Results from the Orman type scheduling of surveillanceand tracking tasks in three sectors of coverage. Requiredsurveillance occupancy = 80%.

0 2 4 6 8 10 12 14 16 18 200

0.2

0.4

0.6

0.8

1

Time (s)

Occ

upan

cy

Region 1Region 2Region 3TrackingTotal Occupancy


this algorithm is that no idle time is left on the radar time line.When the desired surveillance occupancy is less than 100% andthe radar load associated with other functions is less than one,then surveillance tasks are scheduled earlier and the surveil-lance occupancy is increased to the stage where the total radartime line is completely used. Similarly, when the tracking loadis increasing due to a great number of detected targets, the sur-veillance jobs are progressively delayed and the detection per-formance degrades. This could lead to a situation in which nosurveillance is performed and only tracking tasks are executed.

This is in contrast to the Orman type of scheduler, where thehighest priority is given to the tracking function and all thetracking tasks are scheduled as near as possible to their duetimes of execution with the aim of minimizing the number oflate tasks. The surveillance performance is determined by eitherthe radar operator or the mission requirements.

The significance of using the modified version of the MESARscheduling algorithm is that some tasks can be scheduled withgreater delays.

The approaches are compared by analyzing their resultingresource allocation when the multifunction radar faces differentoperational scenarios in respect of the amount of resourcesrequired to meet the performance requirements. Many simula-tions have been performed to validate the algorithms and inves-tigate performance. Two cases are considered here, underload,where not all radar resources are required to carry out the mis-sion, and overload, where there are insufficient resources. Thesehighlight the essential differences and similarities of the twoapproaches. Consider first the case of an underload situation.Figure 2 shows that the Butler algorithm is able to use all avail-able radar resources by scheduling low-priority tasks earlierthan their desired execution time. This behavior is not observedwhen examining the results from the algorithm proposed byOrman, as shown in Figure 3. This is explained by the fact thatonce the performance requirements are met, additionalresources are not used to improve surveillance performance andthe radar has idle periods during its operation. Thus it may beconcluded that, in this case, the radar performance will be supe-rior when the Butler scheme is employed.

In contrast, the results presented in Figures 4 and 5 showthat both schedulers produce similar resource allocation resultswhen operating in overload situations. Here, there are notenough resources to execute all requested radar tasks. Even so,the Orman approach fails to make use of all of the radar timeline and sometimes the occupancy is a little less than 100%. Ingeneral, however, it may be concluded that there is little tochoose between the two in terms of overall performance understressing conditions.

As the multifunction radar is likely to be specified as havingto be able to operate in overload situations, it will be critical thata well-designed scheduler must enable highest priority tasks tobe carried out. This brings into question the assignment of prior-ities. It is now shown that the use of adaptive prioritizationmethods for radar tasks is an efficient way to manage resourceallocation. In particular, the application of fuzzy logic algorithms

developed in [13] to prioritize target tracking and sectors of sur-veillance are considered and compared with fixed prioritizationapproaches. These algorithms are based on expert knowledge andimitate the human decision making in similar situations.

INTELLIGENT PRIORITIZATION USING FUZZY LOGICThe attribution of priority to regions and targets of interest is cen-tral to the eventual performance of the array radar system and tosubsequent mission success. There are a variety of methods thatmay be employed, from simple fixed allocations based on opera-tional experience to more elaborate schemes that attempt to bal-ance competing components that constitute the overalldetermination of priority. For example, the priority for trackingtargets may be evaluated using the decision tree presented inFigure 5. This could be carried out according to information pro-vided by a tracking algorithm, by other sensors, or by other opera-tional modes of the multifunction radar such as high resolution.

[FIG5] Results from the Orman type scheduling of surveillanceand tracking tasks in three sectors of coverage. Requiredsurveillance occupancy = 100%.

[FIG4] Results from the Butler type scheduling of surveillanceand tracking tasks in three sectors of coverage. Requiredsurveillance occupancy = 100%.

0 5 10 15 20 250

0.2

0.4

0.6

0.8

1

Time (s)

Occ

upan

cyRegion 1Region 2Region 3TrackingTotal Occupancy

0 5 10 15 20 250

0.2

0.4

0.6

0.8

1

Time (s)

Occ

upan

cy

Total OccupancyTrackingRegion 3Region 2Region 1



Five different variables provide information used to set the priori-ty level. These are threat, hostility, quality of tracking and relativeposition of the target, and weapon system capabilities of the plat-form. These are termed linguistic variables. The reasons forselecting these variables are explained in the following.

Track quality refers to the accuracy of the predicted positionof the target with respect tothe desired accuracy for thisprediction. For example, ifthe error of the predictedposition is small, the trackquality is said to be high;however, large errors ormissed detections lead topoor quality, and the priority of the target may be increased inorder to improve the prediction of its position.

Hostile is a fuzzy variable related to four concepts: range tothe target, absolute velocity, target identity, and the manner inwhich the target is approaching the radar platform. Thus, thepriority for tracking this target should clearly vary according tothe way the target is approaching the radar platform, itsabsolute velocity, its range, and its identity. These four conceptsare also associated with linguistic variables, which are disposedas nodes of the decision tree and combined to provide the finalvalue for the hostility of a target. Range is the fuzzified value ofthe range of the target. In general, the closer the target, thehigher its priority. Velocity is the fuzzified value of the absolutevelocity of the target. Target identity is the probability of a targetbeing enemy; thus, the higher this probability, the higher thedegree of hostility of the target. Finally, approach is related tothe fuzzified value of the range-rate-to-velocity ratio of a targetrelative to radar platform; high values for range-rate-to-velocityimply a fast approach, leading to high priorities. The variableweapon systems represents the importance of a target in respectof the weapon systems of the radar platform. In order to assessthis importance, three concepts are combined: the identity ofthe target, the operational range of the weapon systems, and theratio between the range rate and the absolute velocity of the tar-get. The range-rate-to-velocity represents the way in which thetarget is approaching the radar platform and also corresponds toa fuzzy variable approach. The Weapons System linguistic vari-able accounts for the operational range at which the weaponsystems can be effectively applied. Targets at long ranges cannotusually be engaged by the weapon system of the radar platform,and therefore their priority is reduced. However, this impor-

tance is gradually increased as the targets move to positions thatare closer to the radar platform; short ranges lead to high targetimportance in respect of weapon systems capabilities. Whencombining the three concepts, different degrees of importancemay be achieved by a target. For example, it is expected thatenemy targets approaching the radar platform fast have greaterpriority than enemy targets moving away when they are detect-ed far away from the radar platform.

Threat is the linguist variable that represents the degree ofthreat of a target according to its trajectory and identity.Trajectory combines four fuzzy variables related to height,maneuver, absolute velocity, and range-rate with respect to thetrajectory on which the target is moving. In general, maneu-vers, low altitudes, high range-rates and high absolute veloci-ties lead to assessing a target as having a high degree of threat.However, even friendly targets may have a degree of threat if

they are moving towardsthe radar platform in situa-tions which may lead tocollisions, for example.

Finally, position is thelinguistic variable whosevalue is given by the combi-nation of the fuzzified val-

ues of the range and azimuth of a target. In most cases, shortranges imply high priority for target tracking as the target maybe situated in high precision tracking areas. Azimuth is a fuzzyvariable that accounts for the existing coherence between theazimuth of a target at its early stages of the tracking and theexpected detection azimuth of threatening or enemy targets.In military applications, this represents the coherence betweenthe position at which the target is detected and the previousknowledge about the environment in which the radar is oper-ating in respect of the distribution of enemy forces.

Fuzzy values are attributed to each variable. Some examplesof the fuzzy values are presented in Table 1. After evaluation ofthese variables according to a set of fuzzy rules, the importance(priority) of the target is determined. While the membership ofeach fuzzy set may take any suitable shape, it is common torestrict the membership functions to triangular, trapezoidal, orbell shaped functions to reduce the computational burdenrequired to determine the degree of membership associated witha particular value of input variable [15]. Here, only triangularand trapezoidal membership functions are used for the fuzzysets associated with the input variables.

A similar methodology is applied to the surveillance func-tion base upon the decision tree presented in Figure 6. In thiscase, the priority of surveillance sectors may be assessedthrough the original priorities attributed to the regions withrespect to the expected tactical scenarios and the informationgathered during the evolution of the actual environments. Thisincludes aspects such as rate of detection of new targets, num-ber of threatening targets, and rate of detection of new threat-ening targets. A set of fuzzy rules enables the evaluation of thepriority of the different sectors considered for surveillance.


FUZZY VARIABLE FUZZY VALUESTRACK QUALITY HIGH AND LOWHOSTILE NONHOSTILE, UNKNOWN AND HOSTILEWEAPON SYSTEMS LOW, MEDIUM AND HIGH CAPABILITYTHREAT VERY LOW, LOW, MEDIUM, HIGH AND VERY HIGHPOSITION CLOSE, MEDIUM AND FARPRIORITY VERY LOW, LOW, MEDIUM LOW, MEDIUM,

MEDIUM HIGH, HIGH, AND VERY HIGH

[TABLE 1] FUZZY VALUES RELATED TO THE MAIN VARIABLESUSED IN THE PRIORITY ASSIGNMENT.

THE EFFICIENT ALLOCATION OF RADARRESOURCES SUCH AS TIME AND ENERGY HAS BEEN THE SUBJECT OF INCREASING

RESEARCH OVER RECENT YEARS.



Apart from the track quality, all the variables are fuzzified inearly stages of the priority evaluation. These fuzzificationsexplain the fact that the universes of discourse, or domains, ofall variables whose membership functions are presented herevary between zero and one. The output of the system is the eval-uated priority of the target under study, which is represented bythe fuzzy variable priority.

The fuzzy representation of the input and output parametersis considered complete when the membership functions, thedomain of each variable, and the number of fuzzy sets (values)are computed [15]. Thus, the determination of the fuzzy if-theninference rules is the next step in the design of the fuzzy logic-based prioritization system. The number of fuzzy rules requiredto assess the value associated with a knot in the decision tree ofFigure 7 is equal to the product of the number of fuzzy valuesthat compose the fuzzy variables linked to that knot. Table 2shows the number of if-then rules used to evaluate the mainfuzzy variables assumed here.

The inferential rules of the fuzzysystem were written on the basis ofintuitive and expert considerations andthen tuned by simulation tests. Theactual number of rules used in theinference system, in some cases, maybe smaller than the required number.This is explained by the fact that partic-ular combinations of fuzzy variablesare very unlikely to be observed in realsystems. The reduced number of rulesdoes not represent a drawback as max-min associations are used by the fuzzyinference system. These associationsensure that the truth of an assertion isnot affected by the number of con-tributing rules but by the degree oftruth of the dominant term. This char-acteristic enables the system to includeadditional rules in the knowledge basewithout concerns about the contribu-tion of other rules. The evaluation ofthe fuzzy rules must follow thesequence proposed on the decisiontree. Thus, the system inputs are fuzzi-fied and successively used to assessother fuzzy variables in cascade to thepoint where the final priority isobtained. It is not especially straight-forward to evaluate how the resultingtarget priority is modified as a functionof the main fuzzy variables by onlyexamining the fuzzy rules. Therefore,graphic representations are a valuabletool to assess the inferential rules.These representations may be obtainedby fixing all the variables but two

involved in the evaluation of the priority. An example of thesesurfaces is presented in Figure 9. The configuration presented inFigure 9 assumes that the values of three variables (track quali-ty, position, and weapon capabilities) are maintained at 0.5, andboth threat and hostile are varied over their respective domains.This configuration may represent a situation in which the targetis located at medium range and has medium importance inrespect of the weapon systems of the radar platform.

[FIG6] Decision tree for sectors of surveillance priority assessment.

Priority

OriginalPriority

Number ofThreatening

Targets

ThreateningTargetsRate

New TargetsRate

[FIG7] Decision tree for targets priority assessment.

Velocity

Priority

ThreatHostile PositionTrack

Quality

Approach

Target ID

Trajectory

Heading

RangeRate

Range

Height

Maneuver RR/V

RR/V

Azimuth

WSCapabilities

WeaponSystems

FUZZY VARIABLE NUMBER OF FUZZY RULESPRIORITY 270HOSTILE 108WEAPON SYSTEMS 16THREAT 36POSITION 6

[TABLE 2] FUZZY VALUES RELATED TO THE MAIN VARIABLESUSED IN THE PRIORITY ASSIGNMENT.


It is observed that, as might be expected, the priority increasesas a consequence of increases in the degrees of threat and hostil-ity of a target. Conversely, low degrees of hostility and threatmaintain the resulting priorities at low levels. Two other areasmay also be identified on this surface. The first is related to thedegree of hostility varying 0.5–1 (medium to very high) and thedegree of threat varying 0–0.5 (very low to medium). The result-ing priority increases as a result of rises in the degree of threator hostility. However, the sensitivity to rises in degree of hostili-ty is greater than the sensitivity to the degree of threat. Thisbehavior is explained by examining situations in which targetswith medium and high probability of being the enemy are mov-ing away from the radar platform, in this case, the higher the dif-ferences in the way the target is moving away from the radar.Thus, the high degree of threat produces a larger effect in thefinal target priority than the lower value of hostility. The second

area corresponds to degrees of threat varying 0.5–1 and low lev-els of hostility. Like the previous situations, the resulting priori-ty increases as a consequence of rises in the degree of threat orhostility. Nonetheless, the behavior is different from the one inthe previous area: the sensitivity to increases in degree of threatis greater than the sensitivity to the degree of hostility. This maybe explained by considering situations where, having low proba-bilities of being the enemy, targets move on threatening trajec-tories towards the radar platform. In this case, the way thetarget is approaching the radar has greater effect in the final pri-ority than its identity. Of course, the manner in which theserelationships are formulated is itself a variable and one in whichexpert input plays a key role. Inevitably there will be a learningprocess during which the rules and relationships are refined as aresult of experience.

Having defined and tuned the fuzzy if-then rules, themethod for prioritizing the relative importance oftracked targets is then validated against test trajecto-ries, for all test trajectories scenarios consist of tar-gets with different identities. The analysis shows thatby knowing the identity of the targets, their priori-ties may vary. This provides valuable information tobe accounted for when deciding how to allocateradar resources in overload situations. Two cases arepresented here for targets moving towards the radarplatform on constant-velocity straight line trajecto-ries. These have been chosen as they represent situa-tions of a high degree of threat where targets may bemoving towards the radar platform to start an attack.In addition, they present the behavior of the methodwhen a variable such as approach is fixed. This helpssimplify the analysis and evaluation of the reasonsfor the results of the prioritization. The system canalso be examined in more complex scenarios whereall variables involved in the prioritization are chang-ing over the simulation.

[FIG9] Graphic representations of the fuzzy rules, considering fixed threevariables: Position, track quality, and weapon systems.

0.8

0.7

0.6

0.5

0.5

0 0

0.4

0.3

0.2

0.20.4

HostileThreat

Prio

rity

0.60.8

1

1

[FIG8] Fuzzy membership functions related to the fuzzy variables representing the degree of hostility and the position of a target.

1

0.8

0.6

Deg

ree

of M

embe

rshi

p

0.4

0.2

0

0 0.1 0.2 0.3 0.4 0.5Hostile

0.6 0.7 0.8 0.9 1

Nonhostile UnknownHostile 1

0.8

0.6

Deg

ree

of M

embe

rshi

p

0.4

0.2

0

0 0.1 0.2 0.3 0.4 0.5Position

0.6 0.7 0.8 0.9 1

Close Medium Far




Figure 10(a) shows the first test trajectory where a targetmoves towards the radar platform on a straight line, having aconstant velocity of 300 m/s. The red circle indicates the originof the trajectory. Three targets are assumed in the analysis. Theyhave the same dynamics and flight height; however, their proba-bilities of being enemy are different as follows: 1 (enemy), 0.5(unknown), and 0.1 (friendly), corresponding to the red, blue,and green curves, respectively. The evolution of the resultingpriorities is shown in Figure 10(b), which shows that, in gener-al, all priorities increase as the targets move towards the radarplatform; and the greater the probability of being enemy, thegreater the resulting priority. Figure 10 also suggests that priori-ties of targets that have unknown identity present a similarbehavior to friendly targets in the early stages of the trajectory.This may be explained by the fact that during that period, therange of the targets is longer than the tactical range of the plat-form weapon systems. This happens until around 60–80 s. From

that instant, as the target is moving close to the boundaries ofthis weapon systems tactical range, the degree of threat of theunknown target is likely to increase. Thus, its priority evolutionhas the similar behavior to the priority evolution of the enemytarget: the closer the unknown target is, the higher and thecloser to the enemy target its priority will be. At short ranges, ifthe identity of the target is still unknown, the target is assumedto be enemy, and its resulting priority is assessed as that.

Figure 11 presents the results of a simulation where targetsare assumed to move on a straight line trajectory with 800 m/sof velocity. The same probabilities of being enemy of the previ-ous analysis are considered. Due to the high velocity and shortranges, the evolution of the priorities is rather different fromthe previous case. During the first few seconds of simulation,both unknown and enemy targets have slightly higher prioritiesthan in the first example. This may be explained by their highvelocities. All target priorities remain fixed until about 30 s,

[FIG10] Resulting priorities for three targets with different probabilities of being enemy, moving on the same trajectory.

00

10

20

30

40

50

60

70

80

90

100

10 20 30 40 50X (km)

Y (

km)

60 70 80 90 100Time (s)

(a) (b)

Prio

rity

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

020 40 60 80 100 120 140 160 180 200

EnemyUnknownFriendly

[FIG11] Resulting priorities for three targets with different probabilities of being enemy, moving on the same trajectory. Target velocity:800 m/s.

00

10

20

30

40

50

60

70

80

90

100

10 20 30 40 50X (km)

Y (

km)

60 70 80 90 100

(a)

Time (s)

(b)

Prio

rity

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

020 40 60 80 100 120 140 160 180 200

EnemyUnknownFriendly


when the target position is getting close to the weapon systemoperational range. Before 30 s, all targets have the maximumpriority possible for the set of characteristics of their dynamics,identity and the capabilities of the weapon systems. Thereafter,the priorities are increased in order to allow the radar platformto face the threat. It isobserved that, from around30–60 s of simulation time,the priority of the unknowntarget presents a high rate ofincrease. The analysis indi-cates that more importanceis progressively given to thistarget which is gradually assumed to be like an enemy target,because its velocity is very high, the target is approaching theradar fast, and its identity is unknown over this period. Fromaround 60–85 s, the unknown target has the highest prioritypossible for the combination of input variables which determineits importance. From 85 s, its priority increases again, reachingits highest at around 100 s, when the target position is withinoperational range of the platform weapon systems and as a con-sequence both enemy and unknown targets have the same pri-ority. Such an unknown approaching target is considered to beof highest importance because of its potential degree of danger,represented by its velocity, the way it is approaching the radarplatform. Like the unknown target, the priorities of both enemyand friendly targets increase from around 30 s, as they are get-ting close to the weapon system operational range. These priori-

ties continue to increase reaching their maximum values notlater than 100 s of simulation, when the position in within theoperational range with a degree of membership of 100%.

The results of the situations examined here suggest that thefuzzy logic approach is an intelligent and valid means for evalu-

ating the priority of targets.By imitating the humandecision-making processand by combining dynamiccharacteristics about radartracking and militaryaspects, such as the ability ofthe weapon systems of the

radar platform to face potential threats, the fuzzy approach mayrepresent an effective and intelligent support for decisionsregarding radar resource management.

FUZZY AND HARD LOGIC FOR PRIORITIZATIONHere, performance is assessed and compared using two prioritiza-tion methods: a fixed priority and a hard logic prioritization. Thefixed priority method is based on the prioritization order typicallyused in radar scheduling. In this analysis, the prioritization orderof Table 3 is used, where tasks related to the tracking functionhave greater importance than tasks related to surveillance.

The prioritization method called hard logic can bedescribed by a set of rules similar to the ones proposed forthe fuzzy logic approach. However, for each operational con-dition, only one rule is fired, determining the priority of theradar task. In this case, the input variables are classified insets using the same labels which described the fuzzy vari-ables. These variables are crisp numbers, which means thatat any time they will only belong to one labeled set. In prac-tice, this method works like the fuzzy logic approach butusing sharp edge membership functions. The main advantageof this method is the reduced computational burden inassigning the priorities of the radar tasks.


PRIORITY RADAR FUNCTION1 SURVEILLANCE (LOWEST PRIORITY)2 TRACK UPDATE3 TRACK INITIATION4 PLOT CONFIRMATION5 TRACK MAINTENANCE (HIGHEST PRIORITY)

[TABLE 3] FIXED PRIORITIZATION ORDER FOR RADAR TASKS.

[FIG12] Resulting priorities for three targets with different probabilities of being enemy, moving on the same trajectory.

SCHEDULING IS AN IMPORTANT SUBPROBLEM OF RADAR RESOURCEMANAGEMENT AND HAS BEEN THE

FOCUS OF INTENSIVE RESEARCH.

00

10

20

30

40

50

60

70

80

90

100

10 20 30 40 50X (km)

Y (

km)

60 70 80 90 100Time (s)

(a) (b)

Prio

rity

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

020 40 60 80 100 120 140 160 180 200



As before, one face of a static multifunction radar consisting ofa four-faced phased array antenna is considered. Thus, a regioncoverage spanning from −45◦ to +45◦ away from the antennabroadside is assumed. In this section, this region is subdivided intothree different sectors of surveillance. For each sector, both frameand dwell times are defined to determine the desired surveillanceperformance. Sector 1 spans 0–45◦ in azimuth and 0–20◦ in eleva-tion. Likewise, sector 2 spans 0◦ to −45◦ in azimuth and 0–20◦ inelevation. These sectors require the highest search rate as targetsmay be detected at short ranges in low elevations. Finally, sector 3extends from the top of sectors 1 and 2 to an elevation of 50◦,spanning from −45◦ to +45◦. Table 4 shows the frame times andbroadside dwell times of this surveillance requirement.

To compensate for the expected decrease in gain when staticsystems are scanning away from array broadside, not only iscoherent integration assumed but also the waveforms used forsurveillance are increased in length. The scan loss is assumed tobe proportional to cos3 θ , where θ is the scan angle off broad-side. The tracking load is represented by a number of enemy tar-gets that, after being inserted in the system, are detected andtracked using Kalman filters.

Examples regarding prioritization of targets is examined usingthe test trajectory presented earlier in Figure 10. The first exampleassumes that three targets are moving towards the radar platformon a straight-line trajectory with a velocity of 300 m/s, as shown inFigure 12(a). Once again the targets have different probabilities ofbeing enemy, being considered enemy (red curves), unknown (bluecurves) and friendly (green curves). The solid lines represent thepriority evolution for the targets when the fuzzy logic approach isused; the dotted lines represent the results using the hard method.

The main difference that can be identified is the soft transi-tion between the different levels of priority when using thefuzzy logic approach. The hard method produces such transi-tions in steps. In general this results in a more efficient deploy-ment of radar resources.

The analysis of the priority evolution when the target is con-sidered to be enemy suggests that both fuzzy and hard resultshave similar trends in respect of priority assignment in spite ofthe different shapes. Similar behavior is observed when com-paring the priority evolution of an unknown target. Asexplained earlier, when the target is far away from the radarplatform, its importance is evaluated as of the friendly target. Inthe case of the hard method, in the early stages of the simula-tion, the unknown target is even considered as important as thefriendly target. However, as the target moves to positions closeto the radar platform, its priority increases at a consequence ofthe target being assessed as dangerous. The priority evolutionis, therefore, similar to the evolution of an enemy target; theunknown is not assessed to be as important as the enemy target

because of the ranges and absolute velocities considered in thisexample. The comparison of the results for the friendly targetsshows that the hard method maintains the target priority con-stant throughout the period of the simulation. This behaviormay be explained by the fact that both range and velocity of thetargets are not sufficient to provoke a change in the prioritylevel at which the target is assessed by the hard method. Thiscomparison shows that both methods may be used to evaluatethe relative priority of the radar targets. Depending on the situ-ation, the priority resulting from the utilization of the fuzzylogic may be greater or not than the priority obtained by theuse of the hard method. At first, there is a tendency to considerthat if two systems execute the same set of tasks, the systemthat assesses these tasks with lower priority should be consid-ered more effective, because fewer resources would be allocatedto execute the tasks. However, this analysis is not always validin radar scheduling, where the resources are demanded by theradar function in order to achieve their performance require-ments. The task priority, therefore, is important for preparingthe set of measurements to be executed by the radar. However,it should also be noted that this does not necessarily mean lessresources. For example in the first 70 s of the scenario, the prior-ity of the enemy target is above that of the hard case and hencewill tend to demand greater radar resource.

CONCLUSIONSIn this article, we have compared the performance of two sched-uling methods and highlighted both differences and similarities.This analysis also indicated that prioritization is a key compo-nent to determining overall performance. A fuzzy logic approachfor prioritizing radar tasks in changing environment conditionswas introduced. Key to the success of this is the prior knowledgebase that sets the fuzzy logic relationships. By assessing the pri-orities of targets and sectors of surveillance according to a set ofrules it is attempted to imitate the human decision-makingprocess such that the resource manager can distribute the radarresources in a more effective way. Results suggest that the fuzzyapproach is a valid means of evaluating the relative importanceof the radar tasks; the resulting priorities have been adapted bythe fuzzy logic prioritization method, according to how the radarsystem perceived the surrounding environment. The analysis ofthese fuzzy logic methods has been further developed by compar-ison with a fixed prioritization order. From the test configurationexamined the nonsmooth transition characteristics associatedwith the hard prioritization suggest that this may lead to verydissimilar and, sometimes, undesirable results.

One aspect, however, must be highlighted. By using the fixedprioritization method the resource manager may not allocate anyresource to low priority sectors, degrading surveillance

SECTOR AZIMUTH COVEEERAGE ELEVATION COVERAGE FRAME TIME BROADSIDE DWELL TIME1 0–45◦ 0–20◦ 2 s 4 ms2 −45–0◦ 0–20◦ 2 s 4 ms3 −45◦ to +45◦ 20–50◦ 3 s 1 ms

[TABLE 4] REQUIREMENTS FOR THE EXAMPLE SURVEILLANCE VOLUME.


substantially or even preventing it from being performed. Similarresults could be found as a result of using the hard and the fuzzylogic approach. Sectors in which no target is detected and inwhich a priori information about the environment suggests thatno threat should be expected may have all surveillance stopped inextreme overload situations. It might be important to determine aminimum surveillance performance to be achieved in all sectorsto avoid dropping the search for new targets in less important sec-tors completely. In real systems, the use of a fuzzy logic methodmay represent a useful support for radar resource managementdecisions. It is also very important that the information in respectof radar resource management is presented to the radar operatorat all times. This knowledge-based information must includeaspects regarding the priorities of the targets and sectors of sur-veillance, how the resources are distributed over the main radarfunctions, and which functions have their performances degradedas a consequence of radar management decisions in resource con-strained scenarios. Therefore, the operator may intervene in theresource allocation as a result of their own assessment of theenvironment or to correct undesirable behavior. Finally, charac-teristics related to growth and maintainability may make thefuzzy logic approaches attractive for military applications.

AUTHORSSergio Miranda is a serving officer of the Brazilian Navy. He hasbeen conducting research at University College London forwhich he has recently been awarded the M.Phil. degree. He isthe coauthor of a number of published conference papers onradar resource management and is the coauthor of two journalpapers on radar resource management that have been submittedfor publication. He has now returned to his naval duties.

Chris Baker is a professor of radar research within theElectronic and Electrical Engineering Department ofUniversity College London. He has been actively engaged inradar system research since 1984 and is the author of morethan one hundred publications. His research interests includecoherent radar techniques, radar signal processing, radar sig-nal interpretation, electronically scanned radar systems, andradar imaging. He is the recipient of the IEE Mountbatten pre-mium (twice), the IEE Institute premium, and is a fellow of theIEE. He is also currently chair of the IEE Radar, Sonar andNavigation systems professional network.

Karl Woodbridge is head of civil radar at the Electronic andElectrical Engineering Department of University CollegeLondon. He obtained a B.Sc. in physics in 1976 and a Ph.D. inmaterials science from the University of Sussex in 1979. He is achartered engineer and a member of the IEE and IOP. Hejoined University College London in 1990 after ten years withPhilips Electronics, latterly as a principal research scientist andproject manager in the semiconductor electronics area.Current radar research activities at University College Londoninclude multifunction, narrowband, bistatic, and netted systemsfor a variety of customers in both civil and defense related areas.The above activities have generated more than 140 journalpapers and conference presentations.

Hugh Griffiths is head of the Department of Electronicand Electrical Engineering at University College London,where he has led a research group since 1988. His researchinterests include radar systems, synthetic aperture systems,and radar signal processing, and he has published more than30 papers and technical articles in the fields of radar, sonar,and antennas. He has given courses on radar and syntheticaperture radar at several international radar conferencesworldwide. In 1996, he received the IEEE AESS FredNathanson Award, and he has also received the IEE Maxwelland Mountbatten Premiums. He is a Fellow of the IEEE, fel-low of the IEE, and in 1997, he was elected to fellowship ofthe Royal Academy of Engineering.

REFERENCES[1] G. Van Keuk and S.S. Blackman, “On phased array tracking and parameter con-trol,” IEEE Trans. Aerosp. Electron. Syst., vol. 29, no. 1, pp. 186–194, Jan. 1993.

[2] E.R. Billam, “Parameter optimisation in phased array,” in Proc. Radar 92, 1992,IEE conf publ., No. 365, pp. 34–37.

[3] J.M. Butler, “Multi-function radar tracking and control,” Ph.D. dissertation,University College London, 1998.

[4] A.J. Orman, C.N. Potts, A.K. Shahani, and A.R. Moore, “Scheduling for a multi-function array radar system,” Euro. J. Operational Res., no. 90, pp. 13–25, 1996.

[5] A. Izquierdo-Fuente and J.R. Casar-Corredera, “Optimal radar pulse schedulingusing a neural network,” in Proc. IEEE Int. Conf. Neural Networks, June–July1994, vol. 7, pp. 4588–4591.

[6] A.P. Stoffel, “Heuristic energy management for active array multifunction radars,”in Proc. 1994 IEEE Nat. Telesystems Conf., San Diego, USA, 1994, pp. 71–74.

[7] D. Strömberg and P. Grahn, “Scheduling of tasks in phased array radar,” inProc. IEEE Int. Symp. Phased Array Systems Technology, Oct. 1996, pp. 318–321.

[8] M.T. Vine, “Fuzzy logic in radar resource management,” in MultifunctionRadar and Sonar Sensor Management Techniques (Ref. No. 2001/173), IEE collo-quium Dig., pp. 5/1–5/4, 2001.

[9] S.L.C. Miranda, C.J. Baker, K. Woodbridge, and H.D. Griffiths, “Phased arrayradar resource management: A comparison of scheduling algorithms,” in Proc.2004 IEEE Radar Conf., Philadelphia, USA, 2004, pp. 79–84.

[10] J.M. Molina Lopez, F.J. Jimenez Rodriguez, and J.R. Casar Corredera, “Fuzzymanagement of surveillance sensors,” in Proc. 37th IEEE Conf. Decision Control,Tampa, USA, 1998, pp. 245–250.

[11] W. Komorniczak, T. Kuczerski, and J.F. Pietrasinski, “The priority assignmentfor detected targets in multi-function radar,” in Proc. 13th Int. Conf. Microwaves,Radar, and Wireless Communications., Poland, 2000, pp. 244–247.

[12] W. Komorniczak, J. Pietrasinski, B. Solaiman, “The data fusion approach tothe priority assignment in the multifunction radar,” in Proc. 14th Int. Conf.Microwaves, Radar and Wireless Communications, Poland, 2002, pp. 627–650.

[13] S.L.C. Miranda, C.J. Baker, K. Woodbridge, and H.D. Griffiths, “Phased arrayradar resource management: A fuzzy logic approach for prioritization of radar tasksand sectors of surveillance,” in Proc. Radar 2004 Conf., Toulouse, France, Oct. 2004.

[14] W.K. Stafford, “Real time control of a multifunction electronically scannedradar (MESAR),” in Proc. IEE Colloquium Real Time Management AdaptiveRadars, June 1990, pp. 7/1–7/5.

[15] G. Bojadziev, and M. Bojadziev, Fuzzy Sets, Fuzzy Logic, Applications.Singapore: World Scientific, 1995.

[16] S. Gopalakrishnan, and M. Caccamo, S. Chi-Sheng, L. Chang-Gun, and L. Sha,“Finite-horizon scheduling of radar dwells with online template construction,” inProc. 25th IEEE Int. Real-Time Systems Symp., 5-8 Dec. 2004, pp. 23–33.

[17] H. Sun-Mog and J. Young-Hun, “Optimal scheduling of track updates inphased array radars,” IEEE Trans. Aerosp. Electron. Syst., vol. 34, no. 3, pp1016–1022, July 1998.

[18] A.G. Huizing and A.A.F. Bloemen, “An efficient scheduling algorithm for amultifunction radar,” in Proc. IEEE Int. Symp. Phased Array Systems Technology,15-18 Oct. 1996, pp. 359–364.

[19] K. Tei-Wei, C. Yung-Sheng, Chin-Fu Kuo, and C. Chang, “Real-time dwellscheduling of component-oriented phased array radars,” IEEE Trans. Comput., vol.54, no. 1, pp. 47–60, Jan. 2005.

[20] W. Komorniczak and J. Pietrasinski, “Selected problems of MFR resourcesmanagement,” in Proc. 3rd Int. Conf. Information Fusion 2000, 10-13 July 2000,vol. 2, pp. WEC1/3–WEC1/8.

[21] D. Stromberg, “Scheduling of track updates in phased array radars,” in Proc.IEEE Nat. Radar Conf., 13-16 May 1996, pp. 214–219.

IEEE SIGNAL PROCESSING MAGAZINE [76] JANUARY 2006IEEE SIGNAL PROCESSING MAGAZINE [76] JANUARY 2006

[SP]


© EYEWIRE Knowledge-Based Resource Management for … · 2015. 7. 19. · ness in radar...

Documents

Transcript of © EYEWIRE Knowledge-Based Resource Management for … · 2015. 7. 19. · ness in radar...