The Synthetic Aperture Sonar Revolution

Transcription

1 21311 Hawthorne Blvd., Suite 300 Torrance, CA l 1555 Wilson Blvd., Suite 320, Arlington, VA (703) The Synthetic Aperture Sonar Revolution Drs. Ralph Chatham Enson Chang Matt Nelson David Marx Angela Putney Kieffer Warman Ken Chick Steve Borchardt Range (m) 370 SAS Image: 10 cm resolution 350 m A zim u th (meters, arbitrary zero point) DTI/rec-1-10/09/2000 This presentation discusses recent results of applying synthetic aperture techniques to sonar. We show a number synthetic aperture sonar (SAS) images processed by Dynamics Technology, Inc. (DTI) from data collected by a variety of different hardware suites. The scientists listed on this page have all contributed to the work. Page 1

2 Ralph Chatham and Dynamics Technology, Inc. DTI is a 24 year old small business that does contract research on hard technical problems, primarily for U.S. government customers 40 employees, ~ 2/3 with Ph.D. level education Synthetic Aperture Sonar (SAS) is one of 4 major business areas Currently working on 9 government & industry SAS research & development contracts Ralph Chatham : U.S. Navy submarine officer (diesel submarines when possible) : DARPA program manager Submarine laser communication, extraction of dissolved O 2 from seawater for UUV use, submarine self defense : Director, Survivability Navy Theater Nuclear Warfare Office : Chief Scientist & Corporate Technical Director, Global Associates Vice president, Dynamics Technology, Inc. SAS business area leader. Along the way Chatham has also been:... One of 120 Astronaut candidate finalists (but he was told in 1980 you are too susceptible to motion sickness; you may go back to sea. ) Chairman, Defense Science Board Task Force on Training and Education Tom Clancy s novel The Hunt for Red October was dedicated to him An occasional public teller of folktales and sea stories DTI/rec-2-10/09/2000 DTI is an American small business specializing in applying physics to signal processing and measurement problems of government and industry. One of our four business areas is the application of synthetic aperture techniques to acoustic data. We have been at this since 1989 and have a number of current contracts for SAS-related research and development. We have five full-timeequivalent Ph.D. level scientists and engineers working in the area. For the most part we work with major hardware manufacturer partners and process data collected by systems they build. We help specify system parameters, monitor the development, participate in the data collection tests and then SAS process collected data. We are also directing the building of a small SAS system as prime contractor under a U.S. government contract. Ralph Chatham is the business area leader in DTI for SAS. The chart lists some, possibly relevant, notes on his background. Page 2

3 Synthetic Aperture Sonar Coherent addition of multiple pings gives: High resolution independent of range & frequency Applicable on scales from meters to tens of kilometers Long range SAS is new 1989: DTI recognized that advances in SAR might correct ocean & motion induced phase errors 1992: Acoustic medium stability experiment 1995: ~200 m range with 5 cm resolution at 50 khz (DARPA) 1997: 1 km range with ~10 cm resolution at 50 khz (DARPA) DARPA/Raytheon 3.2 m 50 khz array 980 Slant Range (m) 1 km Range 10 cm resolution 50 khz SAS system m resolution single ping sonar DTI/rec-3-10/09/2000 Synthetic aperture processing gives high resolution independent of range by creating a long virtual aperture, generated by the coherent addition of multiple pings. The radar world developed synthetic aperture radar (SAR) almost 40 years ago but synthetic aperture sonar (SAS) did not work then because the slow speed of sound allowed uncompensated motion to generate much larger phase errors between pings. Moreover, the ocean is a much less transparent medium that also generates phase errors. Exploiting the full coherence that exists in underwater propagation had to wait until new techniques for data-driven focusing algorithms were developed in SAR world. SAS at long ranges (greater than 20 or so meters) has only been done since 1995, but there are now a number of existence proofs at many scales that it is possible to exploit the coherence in the ocean, and that there is more coherence than had previously been believed. The figure shows fixed aperture (side looking sonar, SLS) vs. SAS images at 1 km. DARPA s Raytheon-built 50 KHz sonar illuminated a sunken PB4-Y2 airplane in Lake Washington. Coherent processing by DTI focused the SAS image on the left. The physical aperture was 3.2 meters and was towed at 2.3 knots. The ~11cm resolution achieved was limited by the number of data channels available in the tow-body (16) and by the low system SNR at this range. Had 16 more channels been available, the resolution would have been ~ 5.5cm. A longer physical receive array could have allowed proportionally faster tow speeds. The synthetic aperture array length was about 300m or 10,000 wavelengths. We will return to this image later. Page 3

4 SAS is real DTI has focused SAS images from wide variety of sonar systems 2.5 cm out to 50 m (180 khz 0.55 m CSS array) 7.5 cm out to 50 m (20 khz 0.55m CSS array) 6cm out to 100m (240 khz 2 m Boeing/Sonatech array) 5 cm out to ~300m (50 khz 1.6 m DARPA array) 10 cm out to 1,000m (50 khz 3.2m DARPA array). and six othe r systems SAS enables order of magnitude performance increases Constant high cross-range resolution to x10 range of current MCM systems Area coverage rate ~same as for current HF side-looking sonars SAS increases pixel rate by >10X for conventional arrays (2-3m long) BUT Real-time long range SAS not yet demonstrated Robustness under wider range of conditions needs to be shown The biggest benefit of SAS will come at long ranges This implies; lower frequencies, lower grazing angles, new propagation paths Caveat: if ray-paths don t illuminate the target, SAS processing won t help DTI/rec-4-10/09/2000 SAS is real, it works in a number of conditions, but there is not yet enough data taken in in large quantities in many environments to give customers confidence that SAS performance will be worth the investment to change their current approaches. Page 4

5 SAS Performance Limits: resolution is independent of range 100 km GLORIA 10 km Sea Mark II ~SLS Technology Limit SAS Limit (Wavelength dependent attenuation) Swath (twice range) 1 km Sea Mark I Klein 50 EG&G 100 DARPA SAS (data) 100 m Klein MBFS EG&G m 100 m 10 m 1 m 10 cm 1 cm Azimuth Resolution Boeing-SAS data CSS SAS(data) DTI/rec-5-10/09/2000 Swath width (twice range) is plotted here against azimuth resolution. Real-aperture systems are plotted in black; SAS data is plotted in color. Resolution with real-aperture systems is bounded by practical array lengths and the wavelength-dependent exponential attenuation of sound in seawater. SAS breaks that barrier by allowing high resolution with longer wavelengths. The longer wavelengths penetrate farther but resolution is maintained (or increased) by by summing many pings coherently thus generating a long synthetic aperture. The ultimate SAS limit also comes from the frequency-dependent attenuation of sea water. All practical sonars (real or synthetic aperture) run up against a wall at some range; very little additional range can be achieved even for enormous increases in system power or gain and that limiting range gets shorter as the wavelength gets shorter. One consequence is that if you want 2.5cm resolution you probably won t get ranges much over a kilometer even with SAS. Page 5

6 Synthetic Aperture Sonar (SAS): why it works SAS gives high cross-range resolution independent of range and frequency, bringing the SAR revolution to sonar. Coherent combination of many pings allows synthesis of larger aperture 10 m 180 m Front view Processing techniques, algorithms & optimizations are key 30 years of SAR research have provided the appropriate tools SLS (side looking sonar) One look: resolution degrades with range DARPA program data, Raytheon 50 KHz Sonar, 10 (-36dB) & 20cm spheres at 180m range. DTI processing DTI-Raytheon SAS Many looks, coherently integrated: resolution is constant with range DARPA/DTI s autofocus algorithms routinely process out 40,000 o f phase error both from uncompensated vehicle motion (over 3m) & from medium fluctuations Real-time computing requirements can be satisfied by common workstations And within the reach of the highest end personal computers on the market Extensive hardware R&D is not required (but new combinations of hardware will be) DTI/rec-6-10/09/2000 Synthetic aperture sonar works by creating a virtual aperture out of multiple pings. The coherently-summed extra looks at more distant targets compensate for the linear degradation of resolution with range that is inherent in real-aperture sonar. The example, from the DARPA program, shows DTI s SAS-processed images of data collected by Raytheon s 50 KHz sonar in The targets were hollow 10-cm (-36dB) & 20-cm spheres at 180 m range. They are hanging in the water column well above the bottom. The key to success in long range SAS is to correct for phase errors arising from uncompensated motion and from medium instabilities. DTI s data-driven focusing algorithms routinely removed 40,000 of otherwise uncompensated phase error over the whole length of up to 10,000 wavelength synthetic apertures. We estimate processing loads for real-time SAS (0 to 1000m range, 10 cm resolution) to be the order of 800 MFLOP using frequency-domain algorithms. Page 6

7 Overview of DTI s SAS Processor 50 khz 3.2m aperture SAS in Dabob Bay. 10 cm resolution. DARPA/Rayteon/DTI program. Water depth <30 m Receive element level data. R = 419 m Create synthetic aperture using multiple returns Range Apply initial motion compensation. Motion corrections can come from inertial sensors or from the data itself. R = 452 m SAS image with no motion or medium compensation 36 m Generate image with frequency-domain all-range focusing algorithm used in seismic and SAR domains, modified for SAS All-range focused image with initial motion compensation Image without SAS: Broadside Beam SLS Apply data-driven auto-focus algorithms to image Display and record Image after autofocus DTI/rec-7-10/09/2000 This chart sketches the process we use to generate SAS images. Element-level time-series data is recorded. Multiple pings are stitched together to form a synthetic aperture. If the platform motion is measured directly, as with an inertial measurement unit, that information is used to decrease the residual phase errors due to motion. If it is not available, as was the case for all the images shown in this brief, then information in the sonar data itself is used to make initial phase corrections for unmeasured motion. An image is then generated by an all-range focusing algorithm, originally developed by the seismic geology and synthetic aperture radar communities, now modified by DTI for the sonar application. Data-driven auto focusing algorithms are then applied to complete the process. The object used to illustrate parts of this process was in Dabob Bay. We don t know what it is. Note that there is a faint shadow behind of the unidentified bow-shaped outcropping. The object is in water of about 30m depth at a range from the sonar of 420m. We show a larger view of this data later. Page 7

8 Imaging Improvements Through SAS Processing Side Looking Sonar (SLS) performance 187 m 5 m 6.2 m Look Direction Slant Range 198 m Initial SAS image SAS image after DTI autofocus processing Hollow air-filled steel spheres (10 and 20cm diameter) above PVC frame and two corner-reflectors. Splitting along range line due to existence of both bottom-bounce and direct path returns DTI/rec-8-10/09/2000 A set of test targets was laid on the bottom in Lake Washington and imaged with the DARPA/Raytheon system. These are the same spheres we showed in a previous chart, but now held a meter above a soft mud bottom on waterfilled PVC posts. We show the the data at various stages of processing. The test spheres focus well in cross range, but are split in the range direction. This, we believe, is due to the existence of multiple paths for the sound to reach the spheres and return. One path is direct. Another includes a bottom bounce. The different paths are different lengths and therefore focus at different ranges in the images above. The brightest returns (at the furthest ranges) are two corner reflectors. They are not split because they can only support reflections that return by the same path as the arrival. The returns from the corner reflectors are much more intense than those from the small spheres (10cm and 20cm), and some of the residual cross-range side-lobe structure can be seen in them. Page 8

9 Unknown Object & Shadow at 230m: SAS vs SLS 215 m SAS 255 m 382 m Range (m ) 245 Range (m) DARPA DTI/rec-9-10/09/ Azimuth (m) Resolution: 11 cm synthetic aperture SAS Azimut h (m) * SLS image as shown is actually over-sampled showing pixels about 1.5 m wide. Data only justifies 2.2m pixels. Here is a object and its shadow as measured in one segment of a SAS run with the 50kHz DARPA/Raytheon hardware. More data was collected in the range direction (~100 to 500m), but due to processing limitations in our 1996-class Pentium-based computers, we only processed a 40m wide strip at a time. SLS 2.2 m* real aperture (SLS) Page 9

10 Dabob Bay Shallow Water 50kHz SAS Data Looking up-slope (top to bottom): water depth at sonar was 30m, shallower at longer ranges. Sand ripples can be seen in the large box in the SAS image. An unknown object about 5 m wide is shown at the same scale with both SLS and SAS resolution. Notice the faint shadow in the SAS image. R = 117 m SAS Range SLS R = 419 m Range SAS image R = 452 m R = 477 m 133 m DARPA Image without SAS: Broadside Beam SLS DTI/rec-10-10/09/2000 Several sets of data were taken by the 50 khz DARPA sonar in Dabob Bay. Here are SAS and SLS images of an area from 117 m to 477 m range. The azimuthal extent of the image is approximately 133 m. The bottom depth at the sonar was approximately 30 m. The depth at the longest range shown is not known, but, since the sonar was pointing in-shore, the depth at the longest range was less than 30 m. Therefore, the images shown represent ranges up to 10 times the water depth. The large box surrounds a field of sand ripples detectable in the SAS image, but not apparent in the SLS view. The small box outlines the data used to generate a blown-up image seen in a previous slide that describes the DTI SAS processor. A clear shadow of an unidentified outcrop is visible in that image despite complications of shallow water propagation. Page 10

11 DTI-Processed SAS Image at 1 kilometer Depth (m) Raytheon 3.2 m 50 khz array Ray angles: -5 o to +5 o Range (km) Sunken Aircraft Imaged Despite Bottom-Bounce Ray-Path 980 m Sound Velocity (m/s) PB4Y at 1 km is illuminated only by reflected ray-paths at the extreme edge of the vertical beamwidth and at the extreme edge of the system design range DTI/rec-11-10/09/2000 DARPA 1001 m We return to the 1000m range, 50 khz, image of the aircraft in Lake Washington, noting this time that it was illuminated by the edge of the 2 vertical beam after a bottom bounce. The return energy also suffered a bottom bounce. There was, nevertheless, sufficient coherence in the signal to focus to near-theoretical resolution. Features of the image include: shadows of the aircraft structure on the wings and shadows of the single vertical rudder on the starboard horizontal stabilizer, and some of the rib structure (in the forward fuselage). This was an accidental discovery; given the predicted ray-paths, we expected there to be nothing to see beyond 500m, but when bright returns were noted in the data at 1 km, we applied SAS processing with the result you see here. Page 11

12 We have a better mousetrap. Whe re is the path to our door? What is different between SAS and current bottom imaging sonars? SAS has much better resolution at much longer ranges SAS has an array-length/range/speed limit Any sonar requires longer wavelengths (~3cm vice 3mm) at longer ranges Long wavelength sonars see more highlights; surfaces are no longer rough Detection performance not yet adequately proven New wavelength regime New geometries (very low grazing angle) New propagation regime Current sonars SAS Customer base is very cautious They want a stock number for SAS systems before they buy Robustly engineered, thoroughly tested, second-generation SAS hardware is not here yet. DTI/rec-12-10/09/2000 SAS still a very new technology. SAS with data-driven focusing is even newer. The technology has yet not moved into commercial or military systems. It will eventually reach them, but there are impediments to their introduction. This chart outlines a few of them. We will elaborate on some of these issues in the rest of this document. Page 12

13 Implications of longer wavelengths DTI/rec-13-10/09/2000 Page 13

14 Imaging a sunken airplane in Lake Washington at 350m m from sonar SAS Image: 10 cm resolution 50 Khz 15m from sonar Range (m) Azimuth (meters, arbitrary zero point) SLS Image: 300 cm resolution 50kHz High frequency (~450kHz) image of the same sunken airplane - unknown sonar Range (m) DARPA Azimuth (m) We show three images of the sunken aircraft in Lake Washington. The upper left is the SAS image. On the lower left is the image made from the same DARPA/Raytheon array used as if it were a conventional side-looking sonar. To the right is the same aircraft imaged by an unknown a high frequency sonar system. Note the different character of the images and that the resolution degrades visibly in the conventional sonar image as the range increases. At first we were puzzled by the extraordinary difference between the very high frequency image taken at 15m from the aircraft and our SAS images collected at 100, 350 and 1000m, particularly the lack of shadows and the ghostly character of the aircraft. (The 350m range image is shown.) We explain the reasons for the differences in the following charts. Page 14

15 SAS Image of PB4Y-2 in Lake Washington (April 1997) 348 m Range 374 m 40 m Ribs? Spacing 60 cm (23") Navy patrol aircraft based upon B-24 design, modified to have a single tail and a pair of AA blisters on aft quarter of fuselage. Wrecked near Seattle in Attempts at salvage tore off port inboard engine and damaged starboard inboard engine. Resting in 50m of water. Starboard wing tip is 4.9m feet above the bottom. This data taken with DARPA-Hughes 3m array and processed as a synthetic aperture image by DTI with 11 cm crossrange resolution, and 3 cm in-range resolution. Vertical angle was about 2 from grazing. Is the 50KHz SAS seeing through the skin into the water-filled interior of the aircraft? DTI/rec-15-10/09/2000 Looking at this image it doesn t look much like a World War II bomber. For example, the wing aspect ratio is too narrow. There are also stripes across the fuselage. Our DARPA sponsor, Dr. Theo Kooij speculated that these stripes were the structural ribs of the aircraft and challenged us to prove or disprove this. Page 15

16 PB4-Y Wing and Wing Spars DTI/rec-16-10/09/2000 The Naval Air Museum, Pensacola, Florida kindly furnished us copies of the plans of this class of aircraft (Navy PB4Y-2). Here you see the wing structure. Note that the wing spars are I-beams. They pass straight through the fuselage. Page 16

17 SAS Measured Spacing of PB4Y-2 Tail Ribbing Range vs. amplitude cuts along aft fuselage One pixel = 3 cm (0.85 pixels/inch) DTI/rec-17-10/09/2000 I (Chatham) initiated a blind test with our (DTI) researchers in Torrance, California. Holding the plans secret from them, I asked them to make amplitude vs range cuts longitudinally along the fuselage and from these to determine the rib spacing. I gave them only one piece of information: the spacing was not uniform. Here are the cuts and the estimates of the spacings for the after fuselage. Each color represents a different azimuthal cut through the image. Page 17

18 Spacing of PB4Y-2 Tail Ribbing Plan dimensions SAS measurement DTI/rec-18-10/09/2000 Here are the plan dimensions superimposed on the amplitude vs. fuselage distance that we measured from the SAS sonar data. The results of our blind test matched the plan dimensions within 2%. It is clear now that Raytheon s 50 khz sonar penetrated the skin of the waterfilled sunken PB4-Y2. If you look back at the SAS image a few pages above, you can see that the ball-turret is clearly imaged at the nose. The stripes visible within the fuselage are the aircraft s ribs. The skin of the wings is not seen, but the wing spars, which pass straight through the fuselage, are clearly imaged, The inboard port engine is missing, having been broken off during a failed salvage attempt. Page 18

19 Theoretical Justification There is theoretical justification (calculated after the fact) for the hypothesis that we are seeing ribs and wing spars. Calculated acoustic transmission through an infinite 0.1" thick aluminum plate immersed in water shows at most 2dB loss at 50 KHz* 30 Transmissio n loss (db) KHz 180 KHz 450 KHz 10 Incidence angle (degrees from perpendicular) *After Junger & Feit, Sound, Structure, and Their Interaction, MIT Press 1986 p 347 and others. Model not to be trusted at grazing incidence since hydrodynamic no-slip condition is ignored DTI/rec-19-10/09/2000 After the fact we calculated that a thin metal skin with water on both sides was virtually transparent to the 50kHz (3cm wavelength) sound used by our SAS. At 450kHz, however, the skin was opaque. This explains both the skeletal appearance of the SAS image and the lack of shadows that are clearly visible in the high frequency (but short range) image. SAS imaged the sunken aircraft at an order of magnitude greater range than conventional side-looking high frequency sonars and did so with very high resolution at all ranges. However, what the lower frequency SAS system saw a different view of the object than did conventional systems at their much shorter ranges but higher frequencies. Page 19

20 Implications of multipath propagation DTI/rec-20-10/09/2000 Page 20

21 Shadow From Direct Path Signal (180 m range) A narrow bundle of rays from -3.5 o to -4 o propagates to the target directly Extended shadow is formed by these rays -25 db 0 db 0 20 rays (-3.5 o to 4 o ) 160 m 1 2 Simple Trapezoidal Plate Target 3 Depth (m) Range.47 m.48 m.98 m Range (km) Target 200 m 5.8 m simulation DTI/rec-21-10/09/2000 At the longer ranges that SAS will enable, shadows may no longer be a useful tool for identification of objects. We modeled the shadow that might result from a vertical trapezoidal plate if a very narrow vertical range of acoustic rays were directed at it. In the next chart we show what a more realistic vertical aperture would illuminate. Page 21

22 Shadow From All Paths (180 m Target) Shadow is shortened by surface reflected paths 160 m Shadow only 0 db Target multipath also spreads into shadow Range 0 20 rays (-25 o to 0 o ) -25 db m 3 Target only Depth (m) db DTI/rec-22-10/09/2000 Range (km) 5.8 m -35 db Simulated data - breaking apart the shadow and target contributions for 25 vertical beamwidth. The shadow disappears and the object image spreads in range into where the shadow would have been when a taller (25 ) vertical aperture is modeled. A number of approaches can be taken to minimize this effect, for example, tailoring of the vertical beam in intensity. Moreover, at longer ranges the phenomenon of mode-stripping (multiple bounces attenuating all but a few more-or-less direct paths) may also help. Nevertheless, the character of images at longer ranges will be different than they are at the shorter ranges of conventional high resolution sonars. If SAS is to be used to its fullest, it must work in this longer range regime. Data must be collected at these ranges and the resulting system performance must be characterized if we are to have confidence in SAS capabilities. This will take a second-generation set of SAS hardware. It is not clear when this will happen. Page 22

23 The SAS Range/Resolution Revolution High resolution sidelooking sonars stop ~ here DARPA/ Raytheon/ DTI SAS 10 cm 5 cm 200 m 400 m 600 m 800 m 1000 m SAS technology will change underwater imaging as SAR did for radar But there are obstacles There is insufficient data in the new long range regimes Longer wavelengths Multipath propagation Low grazing angles Real-time processing not yet done at long ranges Potential users can not afford much research and development Well-engineered second-generation SAS systems do not yet exist There is a revolution coming to a sonar near you. Synthetic aperture processing revolutionized the way that radar imaging is done. The same thing will happen to sonar. There are now clear existence proofs that SAS works and that it can exploit untapped coherence in underwater sound. High resolution sonar imaging at long ranges and high coverage rates is on its way. It is not, however, here quite yet. Page 23