Third-generation naval IRST using the step-and-stare architecture Pierre-Olivier Nouguès∗, Paul Baize, Flavien Roland, Jean-François Olivier, Mathieu Renaudat, Sagem Défense Sécurité, 178 rue de Paris, 91300 Massy, France ABSTRACT With large focal plane arrays now widely available, IR detectors have entered their third generation. Performances have increased dramatically with respect to second-generation, line array detectors, due to the longer integration times afforded. For surveillance systems such as InfraRed Search and Track (IRST), however, operational requirements generally impose a very large field of regard in relation to the instantaneous field of view. This characteristic which has traditionally been obtained through scanning motion for second generation line array detectors must now be rethought to obtain staring operation for 3rd generation FPA, lest motion blurring be incurred. . This paper presents several approaches considered for naval surveillance systems at Sagem Défense Sécurité to tackle this challenge. Three techniques are presented and then compared: fully staring systems, step-and-stare systems, and finally a “modified” step-and-stare system. Keywords: Infrared search and track, IRST, naval, staring, step-and-stare. 1. INTRODUCTION Third generation IRST are now available for naval applications. Based on large, cooled infrared matrices, also known as focal plane arrays (FPA), these systems consist of one or several sensor(s) characterized by a large field of regard, combined with advanced image and signal processing algorithms designed to automatically detect and track threats thanks to their thermal contrast with their background. The new IR sensors achieve greater performances than the previous generation thanks to their staring nature, i.e., to longer integration times (milliseconds rather than microseconds). However, even with increased intrinsic sensor performance, IRST systems are still subject to the tradeoff between detection range and field of view. On one hand the form factor of standard detectors (≅4/3) does not match the shape of the desired surveillance area, i.e., 360° about the horizon. On the other hand, a relatively large coverage in elevation is desirable to address both close surface threats and high elevation airborne threats, compared to the previous generation of naval IRST which were primarily designed to address the sea-skimming missile threat. The IRST designer is hence faced with the problem of achieving the following goals: - maximum detection range on small and/or dim targets - significant elevation coverage about the horizon - limiting the number of IR matrices in the system due to high acquisition and through-life cost In this paper, we present three alternative approaches to the above problem: fully staring, “traditional” step-and-stare and high-velocity, or “modified” step-stare. After a brief description, these solutions are compared with respect to major design choices: optical design, stabilization, and image refresh rate. Performance trade-offs for the various design choices are analyzed. 2. BACKGROUND The first and second generations of naval IRST were designed in the 1980’s and 1990’s to address the shortcomings of ship radar systems, in particular for the detection of sea-skimming missiles in the low-elevation region were the effect of multipath and ducting is the greatest. Using line array scanning infrared detectors, such passive self-protection systems ∗ pierre-olivier.nougues@sagem.com Infrared Technology and Applications XXXIV, edited by Bjørn F. Andresen, Gabor F. Fulop, Paul R. Norton, Proc. of SPIE Vol. 6940, 69401B, (2008) · 0277-786X/08/$18 · doi: 10.1117/12.773953 Proc. of SPIE Vol. 6940 69401B-1 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms are used for panoramic surveillance within a few degrees about the horizon. Larger elevation coverage could not be achieved without sacrificing detection ranges on the smaller, stealthy subsonic missiles. In the 21st century, military navies are increasingly involved in littoral combat rather than purely blue-sea operations. Conflicts frequently involve an asymmetric component, with new threats to warships appearing in the form of FIAC (fast incoming attack craft) and UAV (unmanned aerial vehicles)). These new threats are not well detected by radar systems due to their small RCS (radar cross section), to the lack of radial velocity they often exhibit, and also due to sea clutter close to the own ship. They are therefore ideal candidates for IRST detection, with the added benefit that IRSTs can provide target identification capabilities thanks to the IR imagery they inherently exploit. However, these new threats impose a larger elevation coverage than that typically available on first and second-generation IRST: a field of regard on the order of 10° in elevation is typically required. At the same time, navies still require protection of their high-value assets against new generations of stealthy seaskimming missiles. Due to their high radial velocity, these threats require early detection to be processed by the ship’s combat system, and therefore impose stringent sensitivity requirements: IRSTs should be designed to detect seaskimming missiles as they become visible above the horizon, typically at ranges in excess of 10 nautical miles. In summary, modern IRSTs must survey large sectors (ideally hemispheric) with high detection ranges to meet mission requirements. However, they must do so with a limited number of sensors, since their acquisition price is significant, and more importantly because the number of cooled IR sensors within the system directly impacts its reliability and throughlife cost1. Several alternative designs are described and analyzed in the remainder. 3. FULLY STARING SYSTEMS Systems analyses performed in the US [1] and in the UK [2] have shown that the highest performances of naval IRST are expected from fully staring, highly stabilized sensors with large aperture optics. Sensor resolution has been theoretically shown to be the key factor affecting performance against real backgrounds. This approach requires a large number of imaging matrices with traditional form factor, or alternatively very large matrices with a specific form factor to cover 360° with a high enough resolution. To achieve increased resolution in elevation about the critical horizon region, the use of anamorphic lenses has been selected by some designers [1]. The use of split optics with standard detectors also helps reduce the number of matrices, albeit at the cost of more complex optics [2]. Also, a significant percentage of the FPA’s pixels may be lost due to optical overlap of the fields of view, and in practice their alignment can represent a challenge, especially over a wide range of temperatures. While the above mentioned systems have the highest possible performance (high resolution imaging - 100µrad - class with high refresh rate – at least 10 Hz), their cost will remain high in the years to come for application to operational products. Indeed, the cost of the IRST is more or less proportional to the number of cooled IR cameras required, with their associated optics, and the computing power required to process all video signals. Based on this consideration, some designers [3][4] have attempted to bring the system cost down by (1) limiting the number of sensors to a minimum, and/or (2) suppressing the optical stabilization of the line of sight and resorting to electronic stabilization. The first design choice obviously reduces the resolution of the imaging sensor (in the 1 mrad class) as each matrix must cover a large sector, unless some sort of sector scanning is used, in which case the system falls within the step-and-stare category. This approach may in turn preclude the use of the IRST as an identification sensor, and also hamper its capability to discriminate bona fide targets versus clutter background, as studied in section 6. Finally, the “time-sharing” scanning choice also reduces the image refresh rate available for IRST processing. The key question then becomes “does 1 Cooling machines are life-limited items which require replacement at regular intervals (thousands of hours). Proc. of SPIE Vol. 6940 69401B-2 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms a high image refresh rate compensate the decrease in resolution on the overall IRST system performance?” As is turns out, the trade-off between these conflicting variables is not trivial. Section 6 of this paper provides some insight into the problem. The second design choice, suppressing optical stabilization, results in a decrease of sensor performance due to blurring induced by ship motion during the integration time. The latter must therefore be kept low enough to ideally maintain line of sight motion within the pixel size, which is not conducive to improved sensitivity, since the latter is inversely proportional to the square root of the integration time [2]. 4. THE STEP AND STARE APPROACH Readily available 640x512 IR matrices and ROIC (Read –Out Integrated Circuits) enable snapshot operation at rates in excess of 100 Hz. It is generally admitted that such refresh rates are not required to revisit the surveyed scene in a naval IRST application [1]. Therefore, some form of mechanism enabling to move the line of sight of the imaging sensor over the horizon can help reduce the number of sensors required for 360° coverage. Traditional gimbaled systems are not well suited to obtain fast step-stare operation due to their generally large inertia – a property desirable for stabilization purposes but not for stepping. “Optical multiplexing” using optical invariants is an interesting alternative [3][4], although this approach requires somewhat complex fixed and moving optical components. To simplify even more the system architecture, another solution is to fit an IR camera on a horizontal platform mounted on double gimbals. The horizontal platform is servo controlled to perform stepping motion synchronized with image acquisition. Using COTS components, Sagem DS demonstrated a 4 second recurrence time for a 12° field of view (figure 1). Such image refresh rate is well suited for visual exploitation by an operator in an asymmetric context. However, it is somewhat low for automatic target detection and tracking of airborne threats, which can exhibit significant angular motion between scans. . Figure 1: Sagem DS Panoramic imaging sensor 5. “MODIFIED” STEP-AND-STARE To fully exploit the high frame rate available on matrix IR cameras - typically 100 Hz – for panoramic scanning purposes, a solution for fast stepping must be found. Based on experience gained on 2nd-generation line array IR cameras, Sagem DS designed VAMPIR NG, a naval IRST that uses a focal plane array cooled 3-5 µm IR camera, and small scanners fitted with high quality mirrors to counter the line of motion during the sensor’s integration time (figure 2). Proc. of SPIE Vol. 6940 69401B-3 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms Figure 2: VAMPIR NG IRST The optical design of VAMPIR NG comprises four motorized axes: (1) the training axis, which provides panoramic scanning of the head including IR camera and all sensor optics; (2) the elevation axis, supporting the elevation mirror designed to steer the center of the IR sensor at constant true elevation, regardless of ship roll and pitch motion; (3) the training and elevation scanner axes. In figure 3 below, IR light enters the sensor head through the external window. The first reflection occurs on the mobile elevation mirror whose role is described above. Folding mirror #1 enables the designers to keep the size of the elevation mirror to a minimum over a wide range of elevation angles. A first series of lenses forms a telescope designed to reduce the pupil size prior to use of the scanners. Folding mirror #2 provides for system compactness. Next come the training and elevation scanners, followed by the IR camera objective. These scanners provide a de-scanning action against the continuously scanning of the sensor head, which has components along three axes due to ship roll and pitch. The result is a “freezing” of the FOV direction during the integration time. The IR camera objective is folded thanks to mirror #3, again for sake of system size. SapphireWindow Elevation Mirror IR Detector (hidden) Folding Mirror #3 Telescope Objective Training scanner Folding Mirror #1 Folding mirror #2 (hidden) Elevation scanner Figure 3: Optical design In the rest of this section, we show how the scanners can be servo controlled to obtain the desired “de-scanning” action. Proc. of SPIE Vol. 6940 69401B-4 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms 5.1. Notation 5.1.1. Ownship attitude angles Ownship attitude is defined by three angles in [g] the (East, North, Up) local geographic reference frame. Let K be the heading, T the pitch, and R the roll angles. 5.1.2. Mechanical angles Let [p] be the reference frame attached to the sensor head (SH) fixed part. The training angle C defines the sensor head rotating part position [t] with respect to [p]. The line of sight reference frame is defined as [l]: E is the aiming mirror relative elevation angle S is the line of sight true elevation angle S C is the training scanner angle, S E is the elevation scanner angle. Figure 4 depicts the various reference frames involved. Axis r zt r xt [t] r yt r zms z x M y 2 xr Vs y Vi1 r ym 2 z r xm 1 [m1] E ms Line of Sight r yms x z r zm1 [ms] M r y m1 x 2 z x y Vi2 Pi [m2] r zm 2 4 r xm 2 y R( xt , - Pi ) 2 y r zmc x [mc] r ymc R( zt ,- Pi ) 2 r xmc z r xc r zc z x Vi3 [c] y z Ve IR detector y x r yc Vi5 y z r zs 2 x r r xs 2 Pi [s2] ys 2 S e z 4 y x r x s1 y Vi4 z Pi x 4 r y s1 r [s1] z s1 Sc R( xt , Pi ) 2 C Figure 4: Reference frames Proc. of SPIE Vol. 6940 69401B-5 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms r[p] xp r zp r yp 5.2 Servo loop orders computation The rotating part of the sensor head rotates at a constant rate. This rate is an adjustable parameter, used to compute the training orders thanks to the equation: C = Vc × t . . E can be expressed as a function of the various geometric angles involved : c a ⎞ 1⎛ E = ⎜⎜ arcsin( ) − arctan( ) ⎟⎟ , 2 2 b ⎠ 2⎝ a +b with a = sin T cos C + cos T sin R sin C b = cos T cos R c = sin S To achieve line of sight stabilization, S must be kept constant. The full equations of motion are too lengthy to present here in detail, especially when accounting for offsets in all the angular sensors. However, assuming perfect system geometry, and using approximations for small angles, it can be shown that the scanner rates in training VSc and elevation VSe can be expressed as follows, using notations explained below: VSc = − Vci cos(2 E ) 2 VSe = VE − V Eth 2 To make the computations tractable, we introduce a change of variables based on fictitious angles expressing the position of [l] with respect to [t] (figure 5). [l] Line of Sight T(t,l) = Z(Lth)*Y(Eth)*X(Dth) T(t,l)=Y(2E)*Z(-2Sc)*Y(-2Se) [t] Sensor Head rotating part T(g,l) = Z(-A)*Y(S)*X(D) [g] Geographic T(g,b) = Z(K)*Y(T)*X(R) T(b,t) = Z(C) [p] Ship / SH fixed part Figure 5: Reference frame transition VEth is then the corresponding rate of Eth , theoretical elevation. Using small angles approximations, it can then be shown that: VEth c&(a 2 + b 2 ) − c(aa& '+bb&) a& ' b − b&a with ≈ − 2 2 (a 2 + b 2 ) a 2 + b 2 − c 2 a + b a& ' = (−T& sin T sin R + R& cos T cos R) sin C + T& cos T cos C : b& = −T& sin T cos R − R& cos T sin R c& = S& cos S & during the integration time is not compensated. It can be shown In the above architecture, the variation of image tilt D th that maximum image tilt variation over a revolution is expressed by the relation: max(D& ) ≈ V sin(2 E ) th ci max For a 1 Hz revolution rate, 10° roll/pitch angle, 5 ms integration time, the maximum tilt variation is less than 1 pixel for 99% of the matrix surface area. Therefore, this effect is negligible on IRST performance for all practical applications. Proc. of SPIE Vol. 6940 69401B-6 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms 5.3 Scanning results The above computations are used to drive fast beam-steering training and elevation scanners to counter line of sight motion during the IR camera integration time. Predictive Functional Control (PFC) was used to obtain the required servo performance. A high sampling rate (10 kHz class) was also necessary for the calculations. Figure 6 shows counter-rotation performance in training for a 1 Hz rotation frequency. The actual rate to be countered at scanner level is actually more than 360°/s due to the optics design. In this example, the IR camera operates with a 100 Hz frame rate. The scanner order is set to obtain a several milliseconds of integration time. The position error achieved is 10 µrad (1σ), far smaller than the IR camera IFOV in all IRST applications (100 µrad class). Peak accelerations are in excess of 30 000 rad/s². position (degrees) Order and meas. Counter Rotation Performances 3 0 -3 position error (degrees) 0.002 0 -0.002 time (sec) Figure 6 : Counter rotation performance 6. COMPARISON OF APPROACHES To compare the alternative IRST designs presented in this paper, objective performance criteria must be defined. Since naval IRSTs are designed to detect airborne and surface objects, primarily for self-defence purposes, the main performance criterion is detection range, which translates into response time for the ship’s combat system against the threat, some of which is typically used for object identification or classification. Another important IRST measure of performance is tracking quality. There exist many indicators for this concept, ranging from short track declaration time, to track accuracy and low false track rate. Several detection conditions must be considered to cover the range of conditions encountered by naval IRSTs: simple backgrounds and clutter. Current IRST technology enables airborne threats, even stealthy, to be detected at long enough Proc. of SPIE Vol. 6940 69401B-7 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms ranges that they will be unresolved and will exhibit low contrast irradiance. They will often appear before simple sky backgrounds. Anti-ship sea-skimming missiles, UAV and aircrafts fall within this category. “Close” targets are resolved and typically appear before complex sea or land clutter background. Such targets are FIAC (Fast Incoming Attack Craft) or naval warships. In this section, we analyze the main trade-off separating IRST designs, i.e. the trade-off between resolution and survey rate. We attempt to assess the performance criteria in the two categories of targets mentioned above. 6.1. Long range targets before simple backgrounds The problem of detecting long range stealthy threats consists in finding a weak point target in spatio-temporal uncorrelated noise (typically white noise on uniform background). The main threats that fall within this category are seaskimming missiles just above the horizon, and also low-flying aircraft. There are two basic ways of achieving long range detection in these conditions: (1) increasing the signal-to-noise ratio (SNR) and/or (2) using spatio-temporal noise decorrelation while tracking. 6.1.1 SNR increase: SNR increase can be achieved either by increasing the IR sensor aperture diameter or by increasing the integration time. The noise of an IRST is mainly dependant on aperture size, i.e. noise decreases when aperture size increases. However, the F-number is limited for optical and dewar design reasons, and will often be on the order of 1.5-2 at best. An IRST with a wide field of view will have a much smaller aperture diameter than a narrower field of view IRST with the same F-number. Since noise varies as the inverse of the aperture diameter squared, the SNR of an IRST that has a field of view 3 times larger than another is about 10 times lower. Therefore SNR considerations tend to require small fields of view to achieve sensitivity. Assuming that the optical design of the system is such that wells saturation is not reached, increasing the integration time also enables improving system sensitivity. This, however, is only true for unresolved targets that remain within the same pixel during the integration time. This will be true of remote targets, with a stabilized system, as shown below. Assuming that targets are far enough or with a small enough apparent motion that they appear static between successive samples, then on an unstabilized system the target signal will potentially be spread over several pixels due to own ship motion. In the following example, we attempt to assess the gain that could be made by doubling the integration time for an unstabilized IRST. In the first case, the integration time is T1 , while in the second case it is T2 = 2T1 . Let N1 be the target signal measured during T1 and σ 1 the associated noise. In the second case, we assume that ship motion is such that the target signal is now spread over two pixels. The Point Spread Function (PSF) of the system is not accounted for. As shown on figure 7, the signal received on the two adjacent pixel in the second case is N1 , and the noise associated to T2 is σ 2 = 2σ 1 . In both cases, the detection problem consists in finding a signal in noise. Optimal detection performance is reached using the appropriate adapted filter. We assume that such a filter is available thanks to a priori knowledge of ship motion thanks for instance to inertial sensors. 2N and SNR = N1 + N1 Then SNR1 = N1 = 1 = SNR1 2 σ1 2σ 1 2σ 2 This shows that the best reachable performances are the same as with integration time T1 .This example shows that increasing the integration time is only effective if the target remains within the pixel during integration. Therefore we Proc. of SPIE Vol. 6940 69401B-8 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms could increase pixel size to account for own ship or target motion, but this choice induces a pupil’s diameter diminution as explained above, and therefore a SNR decrease. Ultimately, the only solution is to optically stabilize the line of sight and keep the field of view small if optimal detection range performance is desired in all but benign sea states. Pixel Response mPh integration time 12 = Ti Pixel Response mPh integiation time Ti \T on I p I V Uniform background __ II.I UHH [' : iii'HH Uniform background Li ? II 'n' 2'5 get t p I I It Pixel Number on a specilic line Pixel Number on a specific line Figure 7: Effect of a long integration time on an unstabilized system 6.1.2. Tracking and spatio-temporal techniques The other way to detect stealthy threats at long ranges is either to use spatio-temporal decorrelation of false alarms created on a uniform background while tracking, or by using image processing techniques such as Track-Before-Detect (TBD). Track initialization window sizes mainly depend on the threats’ kinematics, detection noise in plot measurements and angular filtering precision. Using a high frame rate allows reducing track initialization window size by impacting target motion between two successive frames. Track declaration is performed using a criterion that counts the number of detections over a number of frames. This criterion is a trade-off between low false alarm rate and possibility to have detection gaps. Coupled with a CFAR detection algorithm (Constant False Alarm Rate), this high rate tracking allows improving sensitivity by reducing the threshold required to achieve a given false track rate. Simulations performed at Sagem DS have shown that the higher the frame rate is, the better the sensitivity is (but the criteria and the number of hypotheses taken into account are more and more complex). However, the gain reaches a limit, for which initialization windows have reached their smaller value. Threat motion between two frames can be neglected with respect to the others position errors at long ranges. Figure 8 below present the simulation results with the CFAR and high rate tracking approach (the two curves correspond to slightly different initialization rules). SNRUmp is the signal to noise ratio required to create a track with 90% probability, at a false alarm rate of 1 per hour, in 99% of cases. The following assumptions were made, corresponding to long-range sea-skimming missile detection: - maximum target rates in azimuth and elevation: 0.5°/s maximum target acceleration in azimuth and elevation : 0.05°/s². Proc. of SPIE Vol. 6940 69401B-9 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms - unresolved targets uncorrelated white noise background. Variations du SNRUmp pour un ifov de 300µrad. 7 Fenetre imposés Cas normal 6.5 6 SNRUmp 5.5 5 4.5 4 3.5 3 2.5 0 10 1 10 Fréquence en Hz 10 2 Figure 8: Track creation SNR required as a function of survey rate (CFAR) Figure 8 indicates that a 10 Hz survey rate allows using a roughly halved SNR with respect to a 1 Hz survey rate. Increasing the survey rate up to 100 Hz only provides a marginal benefit. This in turns means that a tenfold increase in survey rate only authorizes a doubling of the NEI (Noise Equivalent Irradiance), i.e., a very small increase of the field of view. In addition, a high frame rate allows using spatio-temporal techniques such as TBD without reducing requirements on track declaration delay. Several ways of performing TBD have been implemented and evaluated. One of them is Dynamic Programming. Dynamic Programming is a technique where data are processed over a number of successive frames by testing the spatio-temporal correlation of pixels over trajectories defined by threats motion analysis. Once a number of frames have been analysed, a decision is made about the possible existence of targets. 0-9 09 0.7 0.6 0.6 0.4 0.3 0.2 0-i 0 ID-' 108 I0 10° 1D' lU lU 102 Figure 9: ROC (Pd/Pfa) curves of CFAR and TBD approaches Proc. of SPIE Vol. 6940 69401B-10 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms lU iS' The above figure shows ROC curves obtained using on one hand a CFAR approach (1 frame over 1 second at 1Hz), and on the other hand using Dynamic Programming (10 frames over 1second at 10Hz). Each curve corresponds to the mean of a large amount of different specific trajectories and motions of the target. The red large curve (higher Pd for given Pfa on average) is the result of the dynamic programming performed over 10 frames (1s at 10Hz). The light blue curve is the result of a CFAR algorithm (lower Pd for given Pfa) on 1 frame. The light green curve represents the mathematical “erf” function with the required SNR increase to reach equal performances as the TBD with a CFAR approach at 1Hz. The gain in SNR is about 2.4. Although the simulations presented here are by no means exhaustive, they indicate that TBD or high rate tracking approaches may not provide more than 2.5 times better performances than traditional CFAR methods. In summary, combining results of 6.1.1 and 6.2.2, the benefits of a high survey rate in the case of a remote target (antiship missile, aircraft, helicopter…) over uniform background do not compensate for a lower resolution sensor. 6.2 Close targets before complex backgrounds 6.2.1 Costal Operation When targets are seen before a land background, scene clutter is essentially constant over time, or slowly evolving. Therefore, a high image refresh rate does not provide useful information. Rather, the capability of the system to discriminate between a target and the background will be linked to its resolution, i.e. its capability to facilitate classification of the potential target. 6.2.2. Sea Clutter Background In the presence of sea clutter due to sun glint, the real issue is the frequency distribution of the glint versus that of the targets. Since sun reflection on waves can have a life time of up to a few seconds, track declaration should be delayed by a few seconds to reduce the false track rate. Since the kinematics of targets of interest in sea clutter (FIACs or ships on complex background) are relatively slow, such an approach helps keep the false track rate low without significantly compromising detection range. 6.2.3. Discrimination of false alarms and targets: To discriminate between actual targets and false alarms in cases where object dynamics do not provide the necessary information, IRST systems must resort to classification techniques. These techniques rely on high resolution images. Figure 10 shows the same view of a light inflatable boat and of solar reflection on a wave edge at roughly the same range, with 4 different resolutions (2°x2° survey area). It can be seen that degrading image resolution can altogether preclude discrimination of target and clutter. Also, diluting the target energy over larger pixels can in some cases result in the failure to detect low signature targets and declaration of a false track on sun glint, due to the high energy of such reflections. Proc. of SPIE Vol. 6940 69401B-11 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms Display of a FIAC with a 140µrad ifov Display of a FIAC with a 280µrad ifov 20 Display of a FIAC with a 520µrad ifov 20 60 30 80 40 100 50 15 10 20 25 15 30 120 20 40 60 80 100 120 Display of a solar glint with a 140µrad ifov 60 35 20 40 60 Display of a solar glint with a 280µrad ifov 20 10 40 20 60 30 80 40 100 50 40 20 10 20 30 40 Display of a solar glint with a 520µrad ifov 40 60 80 100 120 60 10 15 20 Display of a solar glint with a 700µrad ifov 5 15 10 20 25 15 30 20 5 5 10 120 . 5 10 40 Display of a FIAC with a 700µrad ifov 5 10 35 20 40 60 40 20 10 20 30 40 5 10 15 20 Figure 10: Effect of resolution on target/clutter discrimination 6.3. Tracking quality: Track quality is a function of sensor position noise. Staring IRST with low resolution will have greater position noise values than fine resolution systems. Unstabilized systems will have even more degraded position measurements. However, high survey rates provide more data input to the trajectory filters. Therefore, Monte-Carlo simulations are required to assess the trade-off between these two conflicting parameters. However, it must be pointed out that tracking accuracy is not an end in itself: it should only be high enough to allow for direct target designation to a weapon system, whether naval gun or short range missile system. 7. CONCLUSION Several approaches to naval IRST design have been presented. • • • Fully staring IRSTs on naval ships require both stabilization and high resolution to achieve optimal performance, in particular for sea-skimming missile detection, and remain therefore expensive due to the number of detectors required. Semi-staring (sector-scanning) IRSTs are more affordable than fully staring IRSTs thanks to the use of fewer IR cameras. They can offer high image refresh rates, however care must be taken not to degrade their resolution excessively, since it has been shown that resolution is more important than survey rate. Modified step-and-stare is a cost-efficient solution allowing for high resolution imaging and high sensitivity, and therefore for both long detection range of small targets and for asymmetric threat detection and identification. 8. REFERENCES [1] E.H. Takken and J.R. Waterman, “Navy DAS Program for SBIRST”, Proc. of SPIE vol. 5406 (2004) [2] Don Manson et al , “Staring Naval Infrared Search and Track Demonstrator”, Proc. of SPIE vol. 5987 (2005) [3] Christophe Grollet et. al , “ARTEMIS: Staring IRST for the FREMM frigate”, Proc. of SPIE vol. 6545 (2007) [4] Zvi Schneider et al., “ELTA’s IRST defense and Self-protection system”, Proc. of SPIE vol. 6542 (2007) Proc. of SPIE Vol. 6940 69401B-12 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/17/2013 Terms of Use: http://spiedl.org/terms