sensors Article
Towards a Cognitive Radar: Canada’s Third-Generation High Frequency Surface Wave Radar (HFSWR) for Surveillance of the 200 Nautical Mile Exclusive Economic Zone Anthony Ponsford 1 , Rick McKerracher 2 , Zhen Ding 3 , Peter Moo 3, * and Derek Yee 1 1 2 3
*
Former Employee of Raytheon Canada Limited, 400 Phillip Street, Waterloo, ON N2L 6R7, Canada;
[email protected] (A.P.);
[email protected] (D.Y.) Raytheon Canada Limited, 400 Phillip Street, Waterloo, ON N2L 6R7, Canada;
[email protected] Radar Sensing and Exploitation Section, Defence Research and Development Canada; Ottawa, ON K1A 0Z4, Canada;
[email protected] Correspondence:
[email protected]; Tel.: +613-998-2879
Received: 21 April 2017; Accepted: 29 June 2017; Published: 7 July 2017
Abstract: Canada’s third-generation HFSWR forms the foundation of a maritime domain awareness system that provides enforcement agencies with real-time persistent surveillance out to and beyond the 200 nautical mile exclusive economic zone (EEZ). Cognitive sense-and-adapt technology and dynamic spectrum management ensures robust and resilient operation in the highly congested High Frequency (HF) band. Dynamic spectrum access enables the system to simultaneously operate on two frequencies on a non-interference and non-protected basis, without impacting other spectrum users. Sense-and-adapt technologies ensure that the system instantaneously switches to a new vacant channel on the detection of another user or unwanted jamming signal. Adaptive signal processing techniques mitigate against electrical noise, interference and clutter. Sense-and-adapt techniques applied at the detector and tracker stages maximize the probability of track initiation whilst minimizing the probability of false or otherwise erroneous track data. Keywords: over-the-horizon radar; HFSWR; cognitive; sense-and-adapt; dynamic spectrum management; dynamic spectrum access; maritime domain awareness; maritime surveillance; MIMO
1. Introduction to Cognitive Radar Early radars were cognitive in the sense that they generally required a person-in-the-loop to optimize performance based on observed data. The goal of modern radar developments has been to replace the person-in-the-loop. This has led to the pursuance of various techniques that adapt the radar and tracking parameters to the mission and sensed conditions. Cognition takes this one step further, adding more and more intelligence to the optimization process. This intelligence may be internally learned or gained from trusted third-party sources. Figure 1 illustrates the core components of a cognitive radar system. The optimizer senses the environment and adapts the transmitter, receiver and processing parameters with the objective of maximizing the wanted signal, whilst simultaneously minimizing the unwanted signal. The correlator isolates and associates time sequential responses with common attributes to maximize the probability of track initiation, whilst minimizing the probability of a missed track or a false track. Both the adaptive radar and cognitive radar utilize feedback from the receiver to the transmitter.
Sensors 2017, 17, 1588; doi:10.3390/s17071588
www.mdpi.com/journal/sensors
Figure 1 illustrates the core components of a cognitive radar system. The optimizer senses the environment and adapts the transmitter, receiver and processing parameters with the objective of maximizing the wanted signal, whilst simultaneously minimizing the unwanted signal. The correlator isolates and associates time sequential responses with common attributes to maximize the Sensors 2017, 17, 1588 2 of 13 probability of track initiation, whilst minimizing the probability of a missed track or a false track. Both the adaptive radar and cognitive radar utilize feedback from the receiver to the transmitter.
Figure 1. Basic components of a cognitive radar. Figure 1. Basic components of a cognitive radar.
However, a cognitive radar also incorporates memory and learning. Intelligence and knowledge However, a cognitive radar also memory and learning. Intelligence and knowledge can be generated by ingesting dataincorporates from other trusted sources. can be generated by ingesting data from other trusted sources. Cognition, as defined in [1], utilizes intelligent signal processing, which builds on learning Cognition, as defined in radar [1], utilizes intelligent signal processing, as which on learning through interactions of the with the surrounding environment, well builds as feedback from the through interactions of the radar with environment, as well feedback fromuses the its receiver to the transmitter, which is the one surrounding of the facilitators of intelligence. A as cognitive radar receiver to the transmitter, which is one of the facilitators of intelligence. A cognitive radar uses its own own data, as well as trusted third-party knowledge, to extend the radar’s capability. For example, if data, as wellisasbeing trusted third-party knowledge, extend the radar’s For for example, if a target a target observed by other meanstoor is known to be capability. of low-value the mission, then is being observed by other means or is known to be of low-value for the mission, then resources resources can be diverted to higher-priority targets. Alternatively, data related to weather conditions, canorbeother diverted to higher-priority targets. Alternatively, related to weather conditions,data, or other environmental data such as wind speed and data direction or ambient temperature can be environmental data such as wind speed and direction or ambient temperature data, can be used to aid used to aid the optimization of the radar. the optimization of the radar. A cognitive radar utilizes the principle of dynamic spectrum access (DSA), which was developed A cognitive radar utilizes principle of dynamic spectrum access radiating (DSA), which was to developed for cognitive radio. DSA is the a spectrum-sharing scheme that allows systems operate as forsecondary cognitive radio. a spectrum-sharing that allows radiating systems operate as users DSA on a isnon-interference andscheme non-protected basis (NIB and NPB).to DSA enables secondary users on a non-interference and non-protected basis (NIB and NPB). DSA enables significant significant improvement in the efficient use of the rf spectrum, and maximizes the number of users improvement in thecompared efficient use the rf spectrum, and maximizes number of users thefeasibility spectrum of of the spectrum toof traditional fixed-spectrum access.the DSA also opens upofthe compared to traditional fixed-spectrum access. DSA also opens up the feasibility of spectrum sharing spectrum sharing between radar and communication users. betweenCognitive radar andtechniques communication users. allow a radar to adapt to a changing threat and clutter environment by Cognitive techniques allow a radar to on adapt to a changing and clutter environment by optimizing performance. Previous work cognitive systems threat and waveform design was presented optimizing performance. Previous work on cognitive systems and waveform design was presented in [2–6]. A pictorial representation of the cognitive cycle is illustrated in Figure 2, where “Sense” in relates [2–6]. Atopictorial representation of theofcognitive is illustrated in Figure 2,ofwhere “Sense” the collecting and processing data, andcycle “Learn” is the understanding the environment relates to on thethe collecting data, and “Learn” the understanding of the environment based sensedand dataprocessing relative toofthe mission profile.isThis data may include information from based on the sensed data relative to the mission profile. This data may include information from independent trusted sources to aid in the detection and tracking process. “Decide” is the determining independent trusted sources aid inalgorithms, the detection tracking “Decide” is the determining of the optimization of thetoradar andand “Act” is theprocess. modification of the radar parameters of the optimization of the radar algorithms, and “Act” is the modification of the radar parameters (waveform, detector, tracker, etc.) to meet the mission objective. The primary objective of the (waveform, tracker, etc.)and to meet mission Theallocation primary objective the cognitive cognitivedetector, system is to detect trackthe targets via objective. the effective of radarof resources, with a system is to detect and track targets via the effective allocation of radar resources, with a secondary objective to confirm the identification of known targets. It can be noted that the cognitive cycle is similar in context to the standard OODA decision cycle: observe, orient, decide, and act.
secondary objective to confirm the identification of known targets. It can be noted that the cognitive Sensors 2017, 17,in 1588 3 ofact. 13 cycle is similar context to the standard OODA decision cycle: observe, orient, decide, and secondary to confirm the identification of known targets. It can be noted that the cognitive Sensors 2017, objective 17, 1588 3 of 13 cycle is similar in context to the standard OODA decision cycle: observe, orient, decide, and act.
Figure 2. The cognitive loop, where “Sense” relates to the collecting and processing of radar data, “Learn” is the understanding of the environment and may include information from trusted sources, Figure Figure 2. 2. The Thecognitive cognitiveloop, loop,where where“Sense” “Sense”relates relatesto tothe thecollecting collectingand andprocessing processingof ofradar radardata, data, “Decide” is the determination of the optimization of the radar, and “Act” is the modification of the “Learn” “Learn” is is the the understanding understanding of ofthe theenvironment environmentand andmay mayinclude includeinformation informationfrom fromtrusted trustedsources, sources, radar parameters meet the mission objectives inofthe environment. “Decide” is the to determination of the optimization theknown radar, and “Act” is the modification of the “Decide” is the determination of the optimization of the radar, and “Act” is the modification of the radar radarparameters parametersto to meet meet the the mission mission objectives objectives in in the the known known environment. environment.
2. Canada’s Third-Generation HFSWR
2. Canada’s Third-Generation HFSWR 2.Canada’s Canada’s Third-Generation third-generation HFSWR HFSWR system has been developed by Raytheon Canada in Canada’s third-generation HFSWR has been developed by on Raytheon Canada in collaboration withthird-generation Defence Research and system Development Canada, based a cognitive sense-andCanada’s HFSWR system has been developed by Raytheon Canada in collaboration with Defence Research and Development Canada, based on a cognitive sense-andadapt architecture [7]. The objective wasDevelopment to utilize knowledge obtained either by sensing the local collaboration with Defence Research and Canada, based on a cognitive sense-and-adapt adapt architecture [7]. The objective was to utilize knowledge obtained either by sensing the local architecture Thetrusted objective was to utilize knowledge obtained by sensing the initiation, local environment or [7]. from third-party sources to maximize the either probability of track environment or from trusted third-party sources to maximize the probability of track initiation, environment or the from trusted third-party to maximize thetracks. probability of track initiation, whilst minimizing probability of false orsources otherwise erroneous whilst minimizing the probability of false or otherwise erroneous tracks. whilst minimizing the probability of false or otherwise erroneous tracks. A second enabling isthat thatititoperates operates a slow-time multiple-input multipleA second enablingfeature featureof ofthe the radar radar is in in a slow-time multiple-input multipleA second enabling feature of the radar is that it operates in a slow-time multiple-input output (MIMO) mode. the radar radarsends sendscoded coded pulses on alternating an alternating output (MIMO) mode.During During operation, operation, the pulses on an basis basis multiple-output (MIMO) mode. During operation, the radar sends coded pulses on an alternating through the two separate transmitting antennas strategically located at either end of the receiver through the two separate transmitting antennas strategically located at either end of the receiver basis through the two separate transmitting antennas strategically located at either end of the receiver array. OnOn receiving, the bothantennas antennas are separated to form a virtual array. receiving, thepulses pulsesreturned returned from from both are separated to form a virtual arrayarray that that array. On receiving, the pulses returned from both antennas are separated to form a virtual array that an equivalent performanceto to a physical aperture ofoftwice thethe length. As illustrated in Figure 3, has has an performance twice length. illustrated in Figure hasequivalent an equivalent performance toaaphysical physical aperture aperture of twice the length. AsAs illustrated in Figure 3, 3, in the vicinity of the indicatedtarget, target, a nominal nominal 33dB improvement in the target-to-clutter ratio ratio was was in the vicinity of the indicated dB improvement in the target-to-clutter in the vicinity of the indicated target, a nominal 3 dB improvement in the target-to-clutter ratio was realized. signal-to-clutterimprovements improvements were gained thethe reduction of the azimuth realized. TheThe signal-to-clutter gainedfrom from reduction of antenna the antenna azimuth realized. The signal-to-clutter improvements were were gained from the reduction of the antenna azimuth beam width by one-half [8]. The azimuth sidelobes visible in the virtual aperture array (VAA) beam width bybyone-half [8].The The azimuth sidelobes in theaperture virtualarray aperture (VAA) beam width one-half [8]. azimuth sidelobes visiblevisible in the virtual (VAA) array response response curve were due to residual amplitude effects, due to the receiving of array cable losses. curve were due to residual amplitude effects, due to the receiving of array cable losses. These effects response curve were due to residual amplitude effects, due to the receiving of array cable losses. These effects can be mitigated with equalization techniques, which were not applied here. caneffects be mitigated equalization techniques, which were notwhich applied here.not applied here. These can bewith mitigated with equalization techniques, were
Figure 3. Comparison of the performance of the virtual aperture array technique relative to standard showing an approximately 3 dB improvement in the target-to-clutter ratio and a reduction of the Figure 3.beam Comparison the performance of of the the virtual array technique relative to standard azimuth widthof byof 50%. Figure 3. Comparison the performance virtualaperture aperture array technique relative to standard showing an approximately 3 dB improvement in the target-to-clutter ratio and a reduction of the showing an approximately 3 dB improvement in the target-to-clutter ratio and a reduction of the azimuth beam width by 50%.
azimuth beam width by 50%.
Sensors 2017, 17, 1588
4 of 13
Sensors 2017, 17, 1588
4 of 13
As illustrated 4, the consists of three an Adaptive, an Optimizer, As illustratedininFigure Figure 4, radar the radar consists ofmajor threesub-systems; major sub-systems; an Adaptive, an and data from trusted third-party sources. For the demonstration system, these auxiliary dataauxiliary sources Optimizer, and data from trusted third-party sources. For the demonstration system, these consisted of aconsisted wide-band monitoring and system a data fusion processor associated data sources of aspectrum wide-band spectrum system monitoring and a data fusionthat processor that the HF radar-derived surface tracks with the known shipping in the area obtained primarily from associated the HF radar-derived surface tracks with the known shipping in the area obtained space-based interception of automatic identification system (AIS) reports. The(AIS) data reports. was combined to primarily from space-based interception of automatic identification system The data provide a comprehensive overviewreal-time of the surface vessel activity. was combined to provide real-time a comprehensive overview of the surface vessel activity.
Figure 4. Canada’s third-generation cognitive HFSWR system consisting of an adaptive HFSWR Figure 4. Canada’s third-generation cognitive HFSWR system consisting of an adaptive HFSWR (Act), (Act), the ingestion and processing of data into the optimizer (Sense), the determination of the the ingestion and processing of data into the optimizer (Sense), the determination of the optimum radar optimum radar processing for the mission (Learn) and the application of optimized processing processing for the mission (Learn) and the application of optimized processing parameters (Decide). parameters (Decide).
The following sections present present examples The following sections examples of of the the intelligent intelligent spectrum spectrum management management that that enables enables robust secondary-users’ operation whilst sharing the spectrum with primary communication users. robust secondary-users’ operation whilst sharing the spectrum with primary communication users. The examples also illustrate how different waveform characteristics can be selected based on The examples also illustrate how different waveform characteristics can be selected based on the the sensed or sensed or known known environment. environment. Examples of of sense-and-adapt sense-and-adapt on on receiving receiving are are presented, presented, which which illustrate illustrate how how unwanted unwanted signals signals Examples are minimized. A new adaptive detector is introduced, which minimizes false detections arising from are minimized. A new adaptive detector is introduced, which minimizes false detections arising from clutter and and other other unwanted unwanted signals. This allows allows the the threshold threshold level level within within the the detector detector to to be be lowered lowered clutter signals. This for improved radar sensitivity. Integrating the tracker within the radar architecture allows the system for improved radar sensitivity. Integrating the tracker within the radar architecture allows the system to move towards a track-before-detect mode by utilizing an adaptive, deferred decision, multi-target to move towards a track-before-detect mode by utilizing an adaptive, deferred decision, multi-target tracker based based on on the the interacting interacting multiple multiple model model (IMM) (IMM) architecture architecture that plot rate rate tracker that handles handles aa high high false false plot without generating significant false tracks. without generating significant false tracks. 2.1. Intelligent Spectrum Utilization 2.1. Intelligent Spectrum Utilization HF Radars must operate on a NIB and NPB with respect to allocated users. This requires that that HF Radars must operate on a NIB and NPB with respect to allocated users. This requires that radar operates with frequency diversity, and that the system automatically selects the best frequency that radar operates with frequency diversity, and that the system automatically selects the best and bandwidth of operation based upon the local spectral measurements and limitations as specified frequency and bandwidth of operation based upon the local spectral measurements and limitations by the radio license. as specified by the radio license.
Sensors 2017, 17, 1588
5 of 13
Sensors 2017, 17, 1588
5 of 13 While operating at a specific frequency and bandwidth, the radar self-monitors for the presence of higher-priority licensed users. Simultaneously, a wideband spectrum monitoring system searches for unoccupied channels thefrequency wider spectrum of operation. The self-monitors spectrum management system While operating at within a specific and bandwidth, the radar for the presence generates a ranking of currently unoccupied channels, on monitoring the missionsystem and searches history of of higher-priority licensed users. Simultaneously, a widebandbased spectrum availability, usingchannels a fuzzywithin logic technique. On the detection of anThe unwanted within the radar for unoccupied the wider spectrum of operation. spectrumsignal management system operating theofspectrum system based instantaneously switches to the of highest-ranked generateschannel, a ranking currently management unoccupied channels, on the mission and history availability, using a fuzzy logic technique. On the of an signal within to theeither radar operate operating unoccupied channel. If no channels aredetection available, theunwanted radar is configurable at a channel, the spectrum management system instantaneously the highest-ranked unoccupied predefined default frequency or cease transmission until switches channelstobecome available [9]. Using the channel. no channels available, the radarthe is configurable to either operate at a predefined default fuzzy logicIftechnique, theare system implements cognitive actions of perception, memory, judgment, frequency or to cease transmission until channels become [9]. Using the fuzzyitlogic and reasoning determine the optimum frequency of available operation. From the above, can technique, be seen that the system implements the cognitive actions of perception, memory, judgment, and reasoning to of the cognitive radar utilizes both narrow-band and wideband channel data to support the actions determine the optimum frequency of operation. From the above, it can be seen that the cognitive perception and memory. radar both narrow-band and wideband channel data support the actions of perception Theutilizes availability of wideband spectrum data offers thetoopportunity to enhance the radar and memory. operation based on cognition. For example, impulses caused by distant lightning events, power lines The availability of wideband spectrum data offers the opportunity to enhance the radar operation or machinery result in the generation of impulsive noise. Impulses, by definition, are wideband based on cognition. For example, impulses caused by distant lightning events, power lines or machinery spectral events. There are several competing factors within the narrow-band channel that challenge result in the generation of impulsive noise. Impulses, by definition, are wideband spectral events. the detection and removal of impulse-contaminated data. The narrow-band channel is also occupied There are several competing factors within the narrow-band channel that challenge the detection with clutter and target information. These signals typically have significant energy that can mask the and removal of impulse-contaminated data. The narrow-band channel is also occupied with clutter noise floor elevation due to the impulse event. In addition, the pulse compression cycle in the HFSWR and target information. These signals typically have significant energy that can mask the noise floor occurs over due several pulse periods, and removal a corrupted pulsecycle necessitates the removal elevation to the impulse event. In the addition, theof pulse compression in the HFSWR occurs of theover entire pulse cycle. several pulse periods, and the removal of a corrupted pulse necessitates the removal of the entire The wideband channel, although occupied by many users, allows for the reliable detection of an pulse cycle. impulse event by detecting noise floor elevation. Asmany mostusers, usersallows are restricted in bandwidth to of a small The wideband channel, although occupied by for the reliable detection an fraction ofevent the wideband theelevation. noise floor increase readily detected between impulse by detectingchannel, noise floor As most usersisare restricted in bandwidth to aoccupied small channels. Tothe determine noisethe floor fromdetected the noise floor,occupied the time-sampled fraction of wideband the channel, noiseand floordiscern increaseusers is readily between channels. wideband datathe is converted to spectral by the noise fast Fourier Thewideband spectral power To determine noise floor and discern power users from floor, thetransform. time-sampled data is of converted spectral power by the fastinFourier spectral powerthat of two timethe two adjacenttotime sweeps is presented Figure transform. 5. It can beThe readily observed the adjacent sweep with sweepsevent is presented Figuretrace) 5. It can observed that the sweep thethe impulse event impulse presentin(blue hadbea readily considerably higher noise floorwith than non-impulsepresent (blue trace) had a considerably higher noise floor than the non-impulse-contaminated sweep. contaminated sweep. This elevation in the noise floor is considerably harder to detect in the narrowThischannel elevation in the noise floor is considerably to detect in the narrow-band channel with band with competing elements of clutter harder and target information.
Power [dB]
competing elements of clutter and target information.
Figure 5. Illustration of the spectral power of two adjacent time sweeps. It can be readily observed that
Figure 5. Illustration of the spectral power of two adjacent time sweeps. It can be readily observed the sweep with the impulse event present (blue trace) had a considerably higher noise floor than the that the sweep with the impulse event present (blue trace) had a considerably higher noise floor than non-impulse-contaminated sweep (red trace). the non-impulse-contaminated sweep (red trace).
Sensors Sensors2017, 2017,17, 17,1588 1588
66of of13 13
A simplified flow diagram for the detection and removal process described above is presented A simplified flowevents diagram the detection and removal receiver. process described above presented in in Figure 6. Impulse are for detected using the wideband These events areisthen removed Figure 6. Impulse events are detected using the wideband receiver. These events are then removed from from the radar data and collected with the narrow-band receiver. Details of the approach are the radar data and collected with the narrow-band receiver. Details of the approach are presented in [9]. presented in [9].
Figure 6. 6. Wideband Widebanddetection detectionof ofimpulse impulseevents eventswith withfeedback feedbackto tonarrow-band narrow-bandtarget targetdetection. detection. Figure
2.2. Intelligence-Based Intelligence-BasedDetection Detection 2.2. Thissection sectionconsiders considers adaptive adaptive variations variations of of constant constant false false alarm alarm rate rate (CFAR) (CFAR) detectors detectors [10–12]. [10–12]. This HF radars operating in the 3–10 MHz region of the spectrum receive significant energy from HF radars operating in the 3–10 MHz region of the spectrum receive significant energy from reflections reflections off the Due ionosphere. Due to the high reflectivity of the the received energy can off the ionosphere. to the high reflectivity of the ionosphere, theionosphere, received energy can be significantly be significantly larger than targets of interest. Thus, the ionospheric returns will generate large larger than targets of interest. Thus, the ionospheric returns will generate a large quantity of aCFAR quantity of CFAR detections, most prevalent along the edges of the returns. Standard CFAR detections, most prevalent along the edges of the returns. Standard CFAR techniques are ineffective techniques are ineffectiveenvironment in the non-homogenous environment within a large, in power in the non-homogenous with a large, abrupt change powerabrupt levels.change With the long levels. With the long integration times necessary for the HF radar and the significant Doppler integration times necessary for the HF radar and the significant Doppler components of the ionosphere, components of the ionosphere, the clutter edges smear significantly theinterference Doppler dimension so that the clutter edges smear significantly in the Doppler dimension so thatinthe environment is the interference environment is non-homogenous. As the ionosphere is not a point target source, the non-homogenous. As the ionosphere is not a point target source, the azimuth component is diffuse, azimuth component is diffuse, whichof leads to aplots significant of Doppler target plots the space. range, which leads to a significant quantity target within quantity the range, andwithin azimuth Doppler andfalse azimuth space.generate The resulting false detections generate significant challenges to The resulting detections significant challenges to maximize the target detection and maximize the target detection and minimize the false track generation in these regions. minimize the false track generation in these regions. Advanced CFAR CFAR techniques techniques can in Advanced can be be successful successful in inmitigating mitigatingthe thequantity quantityofoffalse falsedetections detections non-homogenous environments. AnAn adaptive CFAR process can can be utilized to identify areasareas of highin non-homogenous environments. adaptive CFAR process be utilized to identify of clutter returns in the range–Doppler space to mitigate the false detections. Once the clutter edges high-clutter returns in the range–Doppler space to mitigate the false detections. Once the clutter edges have been been mapped mapped in in the therange–Doppler range–Doppler space, space, the the CFAR CFARparameters parameters can can be beadjusted adjusted to todecrease decrease have the probability of false detections on the edges. Simple techniques such as raising the threshold or the probability of false detections on the edges. Simple techniques such as raising the threshold or adjusting the background test area dimensions are quite effective. adjusting the background test area dimensions are quite effective. Theintelligence-based intelligence-based detector detector uses uses aa hierarchy hierarchy of of CFAR CFARalgorithms algorithmsto todetermine determinethe theoptimal optimal The detector for a given environment and mission. An intelligence-based detector incorporates the detector for a given environment and mission. An intelligence-based detector incorporates the knowledge learned from the environments into the decision process of the hybrid CFAR detection knowledge learned from the environments into the decision process of the hybrid CFAR detection schemes. The The knowledge thethe selection of CFAR detectors, and and is also to adapt schemes. knowledge gained gainedisisused usedinin selection of CFAR detectors, is used also used to the CFAR parameters, including the reference window’s size, position and shape. To control the adapt the CFAR parameters, including the reference window’s size, position and shape. To control number of detections, the knowledge-aided detector adapts the detection threshold to higher levels the number of detections, the knowledge-aided detector adapts the detection threshold to higher in non-homogeneous environments, and to a lower percentage in homogeneous environments. A levels in non-homogeneous environments, and to a lower percentage in homogeneous environments. detector first senses the environments by applying a classifier. The classifier detects the areas of A detector first senses the environments by applying a classifier. The classifier detects the areas clutter and surroundings, the the CFAR CFAR reference reference of clutter andnoise. noise.With Withthe theknowledge knowledgegained gainedfrom from the the local local surroundings, window adapts to best fit the distribution of the local noise/clutter environment. The CFAR setssets the window adapts to best fit the distribution of the local noise/clutter environment. The CFAR threshold so that the rate at which the false alarm occurs is constant. Hybrid CFAR schemes, the threshold so that the rate at which the false alarm occurs is constant. Hybrid CFAR schemes, including order statistics (OS)-CFAR, smallest of CFAR, cell-averaging (CA)-CFAR, and simplified
Sensors 2017, 17, 1588
7 of 13
including order statistics (OS)-CFAR, smallest of CFAR, cell-averaging (CA)-CFAR, and simplified censored CA-CFAR, CA-CFAR, can can then based on on the learned from from the the censored then be be appropriately appropriately applied, applied, based the knowledge knowledge learned classifier of the local environment [13,14]. By deploying this advanced technique, the detector can classifier of the local environment [13,14]. By deploying this advanced technique, the detector can adapt to to the and suppress suppress the the clutter that can can occur adapt the local local clutter clutter surroundings, surroundings, and clutter breakthrough breakthrough that occur when when operating in non-homogenous environments; meanwhile, the detector can adapt to providing more operating in non-homogenous environments; meanwhile, the detector can adapt to providing aa more realistic threshold estimation in homogenous environments, and perform robustly in the presence realistic threshold estimation in homogenous environments, and perform robustly in the presence of of multiple targets. a result,the thehierarchal hierarchaldetector detectorperforms performsoptimally optimallyin in both both homogenous homogenous and and multiple targets. AsAs a result, non-homogenous environments, minimizing false nuisance detections, and increasing the probability non-homogenous environments, minimizing false nuisance detections, and increasing the of detectionof fordetection targets. for targets. probability As illustrated thethe intelligence-based detector is constructed to adapt to specific range, As illustratedininFigure Figure7, 7, intelligence-based detector is constructed to adapt to specific Doppler and beam areas of the detection region. Typically, OS-CFAR types are used for clutter regions, range, Doppler and beam areas of the detection region. Typically, OS-CFAR types are used for clutter and CA-CFAR, for noise-limited regions. The algorithm handles edge effects by freezing thefreezing evaluation regions, and CA-CFAR, for noise-limited regions. The algorithm handles edge effects by the area at the edges of clutter, and moving the test cell from within the evaluation area. This allows for evaluation area at the edges of clutter, and moving the test cell from within the evaluation area. This the detection of targets that are adjacent to strong clutter. This technique produces a more accurate allows for the detection of targets that are adjacent to strong clutter. This technique produces a more estimate of interference levels, compared standard techniques include the strong clutter the accurate estimate of interference levels, tocompared to standardthat techniques that include theinto strong test cells as well as clutter boundaries in the reference window, thereby biasing the detection statistics clutter into the test cells as well as clutter boundaries in the reference window, thereby biasing the of the region. detection statistics of the region.
Figure 7. Construct of the CFAR engine. Figure 7. Construct of the CFAR engine.
Adaptive CFAR techniques minimize the quantity of false detections in non-homogenous Adaptive The CFAR techniques the quantity detections in non-homogenous environments. adaptive CFARminimize process identifies areasofinfalse the range–Doppler space to mitigate environments. The adaptive CFAR process identifies areas in the range–Doppler space to mitigate the the false detection problem. Once the clutter edges have been mapped in the range–Doppler space, falseCFAR detection problem. Once thetoclutter edges have been of mapped in the range–Doppler space, the parameters are adjusted decrease the probability false detections on the edges. Simple the CFAR parameters are adjusted to decrease the probability of false detections on the edges. techniques such as raising the threshold or adjusting the background test area dimensions are quite Simple techniques such as raising theextended thresholdthe or adjusting themodel background dimensions are effective. In our solution, we have two-clutter in [15]test to area a three-clutter-type quite effective. In our solution, wemediumhave extended the two-clutter in [15] to a three-clutter-type model, to identify areas of low-, and high-clutter for model each range–Doppler plane of each model, to identify areas of low-, mediumand high-clutter for each range–Doppler plane of each azimuth beam. Within each of these zones, the CFAR parameters are specified, with the overall goal azimuth beam. Within each of these zones, the CFAR parameters are specified, with the overall goal to to maximize the target detection probabilities in the low- and medium-clutter areas, and minimize maximize the target detection probabilities false detections in the high-clutter regions. in the low- and medium-clutter areas, and minimize false detections in the high-clutter regions. Figure 8a presents a typical range–Doppler plane mapped to color (power), whereas Figure 8b Figure 8a presents a typical range–Doppler plane edges mapped to color (power), whereas 8b shows the same data with the clutter areas and clutter overlaid with a red and whiteFigure overlay. shows the same data with the clutter areas and clutter edges overlaid with a red and white overlay. Within this area, the CFAR parameters are adapted to minimize the detection likelihood due to clutter Within this area, the CFAR parameters are adapted to minimize the detection likelihood due to
Sensors 2017, 17, 1588 Sensors Sensors 2017, 2017, 17, 17, 1588 1588
8 of 13 88 of of 13 13
clutter edges. A comparison of the performance between a non-adaptive CFAR and our adaptive CFAR isA shown in Figure 9. Itperformance can be observed thataaanon-adaptive balanced detection process was achieved edges. comparison of between CFAR our CFAR is edges. A comparison of the the performance between non-adaptive CFAR and and our adaptive adaptive CFARwith is shown 9. be that detection process was with maximum sensitivity in non-clutter and a reduced sensitivity the clutter shown in in Figure Figure 9. It It can can be observed observed(homogenous) that aa balanced balancedregions detection process was achieved achieved withatmaximum maximum sensitivity (homogenous) regions and aa reduced sensitivity at clutter edge. edge. In the in example presented, our adaptive CFAR decreased the total CFAR count by approximately sensitivity in non-clutter non-clutter (homogenous) regions and reduced sensitivity at the the clutter edge. In In the the example presented, our adaptive CFAR decreased the total CFAR count by approximately 90%, while 90%, while maintaining valid detections on all targets. The dramatic CFAR-hit reduction results for example presented, our adaptive CFAR decreased the total CFAR count by approximately 90%, while maintaining valid detections on all targets. The dramatic CFAR-hit reduction results for significantly significantly processing loads on the tracker allowed for more complex tracking techniques maintainingreduced valid detections on all targets. The dramatic CFAR-hit reduction results for significantly processing loads on the tracker for tracking techniques to utilized, toreduced be utilized, resulting of less false complex tracks and a more consistent tracking reduced processing loadsin onthe thegeneration tracker allowed allowed for more more complex tracking techniques to be be utilized,of resulting in the generation of less false tracks and a more consistent tracking of legitimate targets. legitimate resulting targets. in the generation of less false tracks and a more consistent tracking of legitimate targets.
Figure (a) Typical range–Doppler plane, Figure high-clutterareas, areas,and andedges edgesidentified identifiedwith withaa ared red Figure8.8. 8.(a) (a)Typical Typicalrange–Doppler range–Dopplerplane, plane, with with (b) (b) high-clutter high-clutter areas, and edges identified with red and white overlay. and white overlay. and white overlay.
Figure 9. Range–Doppler plane with with baseline CFAR CFAR detectionoverlay overlay (left)and and adaptiveCFAR CFAR Figure Figure 9. 9. Range–Doppler Range–Doppler plane plane with baseline baseline CFAR detection detection overlay (left) (left) and adaptive adaptive CFAR detections overlay (right) showing an approximately 90% reduction in CFAR detection with detections detections overlay overlay (right) (right) showing showing an an approximately approximately 90% 90% reduction reduction in in CFAR CFAR detection detection with with adaptive adaptive adaptive CFAR. CFAR. CFAR.
Sensors 2017, 17, 1588
9 of 13
2.3. Intelligence-Based Tracking A general overview of the tracker used within the radar system can be found in [16]. The application of cognition introduces the use of external knowledge and intelligence into the tracking process. Intelligence-based tracking is used to maximize the probability of a wanted track, whilst minimizing the probability of an unwanted or otherwise erroneous track.
• • • • •
Intelligence based tracking can be implemented using one or all the following techniques: Feature aided tracker (FAT) Special track processing techniques Enhanced tracking filters Enhanced track classification algorithms
FAT is an emerging technology that has been specifically designed for challenging tracking scenarios, e.g., the tracking of weakly detected targets in heavy clutter. FAT begins with the identification of existing and new target features. A feature extraction algorithm is then designed to extract features from a continuous string of signature signals that are commensurate with the targets of interest being tracked over time. Potentially discriminating target features are power, radar cross section (RCS) or the size of the target, for instance. Kinematic and target features are then jointly exploited in a probabilistic manner for simultaneous tracking and identification or classification. Target features and the kinematic properties of the target (in track) jointly determine the weights for data association, which not only reduces miscorrelation, but as a result, greatly improves track continuity. Thus, in demanding tracking scenarios, FAT renders a significant improvement over conventional tracking schemes that are based solely on the kinematic properties of the target in track. Examples of the application of FAT techniques can be found in [17,18]. To further improve performance, the tracker also incorporates special processing techniques to reduce the occurrence of false tracks and seduced tracks in a high-clutter environment. Some of these techniques are described below. 2.4. Adaptive Track Promotion Logic Adaptive track promotion logic defines a track lifeline as potential, tentative, confirmed and deleted. Details of the algorithm can be found in [16]. As an example, a typical fixed-track promotion logic may be defined as 3-4-6. This logic states that a tentative track (that is newly formed with two plots) will be deleted if it has three consecutive misses. A tentative track will be promoted to a confirmed track if there are four accumulated associations. If a confirmed track fails to associate plots in six consecutive updates (i.e., six consecutive misses), it will be deleted. For comprehensiveness, the logic can be written as 2-3-4-6, where the number 2 is the number of misses to delete a potential track. However, the number 2 is never changed (in our scheme); thus, the more concise form of 3-4-6 is typically adopted to describe the track promotion logic. A fixed track promotion logic is adequate for a homogeneous environment, but less so for a heterogeneous environment. For instance, in a “clean” environment, it would be prudent to select a promotion logic that resulted in a timely track confirmation. However, in an environment of heavy clutter, the higher probability of a false track may preclude such a configuration for the track promotion logic. Thus, it is desirable to employ an adaptive promotion logic that is not fixed, but varies according to the local environment. When adaptive promotion logic is utilized, the track promotion logic is adjusted based on the clutter distribution. Adaptive track promotion logic relies on a plot concentration map to monitor the plot activity within the instrumented range of the radar, and to identify areas with excessive plot activity. In areas of excessive plot activity, the likelihood of false track confirmation from a sequence of extraneous plots is high; thus, the parameters governing the track promotion are locally adapted within these areas. This adaptation significantly reduces the rate of confirming false tentative tracks in areas of excessive clutter.
Sensors 2017, 17, 1588
10 of 13
Sensors 2017, 17, 1588
10 of 13
Other Other parameters, parameters, such such as as the the gate gate threshold threshold for for ellipsoidal ellipsoidal gating, gating, may may also also be be adapted, adapted, and and the the adaptation may even be augmented by knowledge of the local infrastructure or terrain, such adaptation may even be augmented by knowledge of the local infrastructure or terrain, such as as in in map-aided map-aided processing. processing. Figure Figure 10 10 shows shows range range vs. vs. time time track track data data between between aa tracker tracker employing employing fixed fixed promotion promotion logic logic (red tracks) and a tracker using adaptive promotion logic (blue tracks). Raw input (red tracks) and a tracker using adaptive promotion logic (blue tracks). Raw input plot plot data data was was collected from an operational HFSWR system. Across 4 hours of data, it is apparent that the tracker collected from an operational HFSWR system. Across 4 hours of data, it is apparent that the tracker using logic generated significantly moremore false false trackstracks (at near-range) than thethan tracker using aafixed fixedpromotion promotion logic generated significantly (at near-range) the which its promotion logic. Adaptive promotion logic dynamically adapts theadapts gate threshold trackeradapted which adapted its promotion logic. Adaptive promotion logic dynamically the gate and the parameters governing track promotion toaccording local clutter and consequently, threshold and the parameters governing trackaccording promotion toconditions, local clutter conditions, and renders a track picture that is effectively free of false tracks, while demonstrating negligible impacts consequently, renders a track picture that is effectively free of false tracks, while demonstrating on the tracking of genuine targets. As adaptive promotion provides an effective negligible impacts on the tracking of shown genuineabove, targets. As shown above, logic adaptive promotion logic mechanism to mitigate false tracks due to strong clutter with non-stationary statistics, and thus, provides an effective mechanism to mitigate false tracks due to strong clutter with non-stationary it has beenand implemented the implemented tracker for theinthird-generation HFSWR system. statistics, thus, it hasin been the tracker for the third-generation HFSWR system.
Figure 10. Range Rangevs. vs.time time track overlay comparing fixed adaptive promotion Figure 10. track overlay comparing fixed (red)(red) and and adaptive (blue)(blue) promotion logic logic tracking. tracking.
2.5. Map-Aided Processing 2.5. Map-Aided Processing A infrastructure cancan alsoalso be combined withwith the plot concentration map A map map of ofthe theterrain terrainororlocal local infrastructure be combined the plot concentration to quickly identify suspect plots/tracks. For example, knowledge of the coastline could be used to map to quickly identify suspect plots/tracks. For example, knowledge of the coastline could be used suppress land-based tracks, whereas knowledge aboutabout the local could allow tracker to suppress land-based tracks, whereas knowledge the infrastructure local infrastructure couldthe allow the to quickly identify false plots/tracks caused by active offshore structures. This suppression can also tracker to quickly identify false plots/tracks caused by active offshore structures. This suppression be radar Map-aided knowledge the localabout surroundings, canused alsoinbe usedspectra. in radar spectra. processing Map-aidedleverages processing leveragesabout knowledge the local and can dramatically the rate reduce of falsethe tracks clutter. surroundings, and canreduce dramatically ratecaused of falseby tracks caused by clutter. Intelligence-based the decision problems related to track confirmation and track Intelligence-basedtracking trackingaddresses addresses the decision problems related to track confirmation and termination. In general, such decisions will be a function of theof probability of target detection, local track termination. In general, such decisions will be a function the probability of target detection,
Sensors 2017, 17, 1588
11 of 13
false alarm density, sequence of associated plots, and the lifetime of the track in question. For a better utilization of all the available information, track confidence can also be used to establish/terminate a track. Track confidence will typically require sensible estimates for the local false alarm density and the ever-changing target detection probability. Another simpler approach would be to merely assign a higher priority to longer, more consistent tracks. The rationale is that by assigning a higher priority to longer, more consistent tracks, the likelihood that a real target plot will be associated with a false track will tend to be reduced. This technique termed “processing prioritization” promotes track continuity for established tracks. Established tracks are maintained through a tracking filter, as discussed in the following section. 2.6. Enhanced Tracking Filters There is a plethora of tracking filters, the most common (in practice) being the alpha − beta filter and the Kalman filter. These single-model filters cannot provide a sensible compromise between two conflicting requirements: significant noise reduction during uniform motion, and a fast response during a maneuver. Consequently, multiple model filters were considered to better handle the tracking of maneuvering targets. A notable development is the IMM filter, which has emerged as the most cost-effective approach for the tracking of maneuvering targets via multiple model filtering. This approach is very efficient when handling a large-scale target situation consisting of many maneuvering targets, and it has been shown to outperform both the alpha − beta and the Kalman filters in such an environment [19]. 2.7. Enhanced Track Classification Algorithms This has been implemented through a classifier-aided tracker based on the support vector machine (SVM) technique for target track classification [20]. By performing plot/track classification, the classifier aids the tracker in selecting the appropriate plot to track assignment. The classifier-aided tracker, as shown in [18], yields considerable improvements in the presence of high-clutter sources, with respect to conventional tracking techniques. 3. Results As discussed previously, the use of an intelligence-based CFAR detector can result in a significant reduction in the number of false detections arising from the presence of unwanted clutter. This reduces the processing load on the tracker, and allows more resources to be allocated to advanced tracking techniques, with the result of maximizing the probability of tracking, whilst minimizing the probability of false tracks. As an example, Figure 11 presents a range vs. time track picture overlay for a 6 hour recording based on raw radar data collected from a HFSWR system located near Halifax, Nova Scotia. The red tracks were obtained through the baseline CFAR detector whereas the blue tracks were obtained through the adaptive CFAR detector. The baseline detector employed the same threshold and reference window for each test cell, whereas the adaptive CFAR adjusted the parameterization (e.g., threshold and background test area) according to whether the test cell resided in an area of low-, medium- or high-clutter. Both sets of data used the same tracker. Owing to the adaptive CFAR detector, it can be seen that several long-range targets enjoyed an extended track life and earlier detection (circles above a range of 250 km), and that a number of false tracks arising from ionospheric clutter were completely eliminated from the track picture (near a range of 120 km).
Sensors 2017, 17, 1588 Sensors 2017, 17, 1588
12 of 13 12 of 13
Figure 11. Overlay of range vs. time track picture via baseline and adaptive CFAR detectors. It can be Figure 11. that Overlay of range vs. timedetector track picture via baseline CFARtracks detectors. It (circles can be observed the adaptive CFAR extended the trackand lifeadaptive and initiated earlier observed that the adaptive CFAR detector extended the track life and initiated tracks earlier (circles above a range of 250 km), and that a number of false tracks arising from ionospheric clutter or other above a range a number of false tracks from ionospheric clutter or other sources (at 120ofto250 140km), km) and werethat completely eliminated from arising the track picture. sources (at 120 to 140 km) were completely eliminated from the track picture.
4. Conclusions 4. Conclusions This paper presented an overview of Canada’s third-generation HFSWR that has been developed This paper presented an overview of Canada’s third-generation HFSWR that has been to form the foundation layer of a maritime domain awareness system to provide enforcement agencies developed to form the foundation layer of a maritime domain awareness system to provide with real-time persistent surveillance out to and beyond the 200 nautical mile EEZ. It has been shown enforcement agencies with real-time persistent surveillance out to and beyond the 200 nautical mile that cognitive sense-and-adapt technology and dynamic spectrum management ensure robust and EEZ. It has been shown that cognitive sense-and-adapt technology and dynamic spectrum resilient operation within the highly congested HF band. It has been shown that dynamic spectrum management ensure robust and resilient operation within the highly congested HF band. It has been access enables the system to operate on a NIB and NPB without impacting other spectrum users. shown that dynamic spectrum access enables the system to operate on a NIB and NPB without The paper has discussed adaptive signal processing techniques that mitigate against electrical noise, impacting other spectrum users. The paper has discussed adaptive signal processing techniques that interference and clutter. Sense-and-adapt techniques applied at the CFAR detector and tracker stages mitigate against electrical noise, interference and clutter. Sense-and-adapt techniques applied at the have been shown to dramatically reduce the number of false detections associated with clutter, CFAR detector and tracker stages have been shown to dramatically reduce the number of false and thereby significantly reduce the processing load on the tracker. This allows more advance detections associated with clutter, and thereby significantly reduce the processing load on the tracker. tracking techniques to be deployed, such that the system maximizes the probability of tracking, This allows more advance tracking techniques to be deployed, such that the system maximizes the whilst minimizing the probability of false or otherwise erroneous track data. probability of tracking, whilst minimizing the probability of false or otherwise erroneous track data. Acknowledgments: Third-Generation HFSWR was developed by Raytheon Canada under a project funded by Acknowledgments: Third-Generation HFSWR was developed by Raytheon Canada under a project funded by Defence Research and Development Canada. Defence Research and Development Canada. Author Contributions: Peter Moo and Zhen Ding established the performance requirements for Third-Generation HFSWR. Anthony Ponsford, Rick McKerracher, and Derek Yee contributed to the design and development of the Author Contributions: Peter Moo and Zhen Ding established the performance requirements for Third-
HFSWR. All authors contributed to writing the paper. Generation HFSWR. Anthony Ponsford, Rick McKerracher, and Derek Yee contributed to the design and Conflicts of Interest: The authors declare no conflict of interest. development of the HFSWR. All authors contributed to writing the paper.
Conflicts of Interest: The authors declare no conflict of interest. References 1. Guerci, J.R. Cognitive radar: The next radar wave? Microw. J. 2011, 54, 22–36. References 2. 1. 3.
Haykin, S. Cognitive Dynamic Systems: Perception-Action Cycle, Radar and Radio; Cambridge University Press: Guerci, J.R. Cognitive Cambridge, UK, 2012.radar: The next radar wave? Microw. J. 2011, 54, 22–36. Lane, R.O. Cognitive Radar: The Knowledge-Aided Fully Adaptive Approach. Aeronaut. J. 2011, 115, 390. [CrossRef]
Sensors 2017, 17, 1588
4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.
19. 20.
13 of 13
Aubry, V.; Carotenuto, A.; Maio, A.D. Forcing Multiple Spectral Compatibility Constraints in Radar Waveforms. IEEE Signal Process. Lett. 2016, 23, 483–487. [CrossRef] Aubry, A.; Maio, A.D.; Farina, A.; Wicks, M. Knowledge-aided (potentially cognitive) transmit signal and receive filter design in signal-dependent clutter. IEEE Trans. Aerosp. Electron. Syst. 2013, 49, 93–117. [CrossRef] Aubry, A.; Maio, A.D.; Naghsh, M.M. Optimizing Radar Waveform and Doppler Filter Bank via Generalized Fractional Programming. IEEE J. Sel. Top. Signal Process. 2015, 9, 1387–1399. [CrossRef] Moo, P.; Ponsford, A.M.; DiFilippo, D.; McKerracher, R.; Kashyap, N.; Allard, Y. Canada’s Third Generation High Frequency Surface Wave Radar System. J. Ocean Technol. 2015, 10, 21–28. Wang, J.; Ponsford, A.M.; Wang, E.; McKerracher, R. Virtual Antenna Extension for Sampled Aperture Arrays. U.S. Patent 9638793, 2 March 2017. McKerracher, R.; Moo, P.; Ponsford, A.M. Spectrum Utilization: Sense and Adapt: Operation on a Non-Interference and Non-Protected Basis. IEEE Spec. Issue HF Radar Aerosp. Electron. Sys. Mag. 2017, in press. Gandhi, P.P.; Kassam, S.A. Analysis of CFAR processors in homogeneous background. IEEE Trans. Aerosp. Electron. Syst. 1988, 24, 427–445. [CrossRef] Farrouki, A.; Barkat, M. Automatic censoring CFAR detector based on ordered data variability for nonhomogeneous environments. IEEE Proc. Radar Sonar Navig. 2005, 152, 43–51. [CrossRef] Shor, M.; Levanon, N. Performances of order statistics CFAR. IEEE Trans. Aerosp. Electron. Syst. 1991, 27, 214–224. [CrossRef] Wang, E.; Wang, J.; Ponsford, A.M. An Adaptive Hierarchal CFAR for Optimal Target Detection in Mixed Clutter Environments. IEEE Radar Conf. 2011, 543–547. [CrossRef] Wang, E.; Wang, J.; Ponsford, A.M. Knowledge Aided Detector. U.S. Patent Application 13/674,563, 13 October 2015. Chen, B.; Varshney, P.K.; Michels, H. Adaptive CFAR detection for clutter-edge heterogeneity using Bayesian inference. IEEE Trans. Aerosp. Electron. Syst. 2003, 39, 1462–1470. [CrossRef] Ding, Z.; Sinha, A.; Kirubarajan, T.; Farooq, M. A Track Quality Based Multi-Target Tracker. U.S. Patent Application 11/756,913, 1 December 2009. Vivone, G.; Braca, P.; Horstmann, J. Knowledge-Based Multi-Target Ship Tracking for HF Surface Wave Radar Systems. IEEE Trans. Geosci. Remote Sens. 2015, 53, 3931–3949. [CrossRef] Sergey, L.; Hubbard, O.; Ding, Z.; Ghadaki, Z.; Wang, J.; Ponsford, A.M. Advanced Mitigating Techniques to Remove the Effects of Wind Turbines and Wind Farms on Primary Surveillance Radars. IEEE Radar Conf. 2008, 1–6. [CrossRef] Bar-Shalom, Y.; Li, X.R.; Kirubarajan, T. Estimation, Tracking and Navigation: Theory, Algorithms and Software; Wiley: New York, NY, USA, 2001. Dizaji, R.M.; Ghadaki, H. Classification System for Radar and Sonar Applications. U.S. Patent Application 11/401,097, 28 July 2009. © 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
