Test Results from a Multi-Frequency Bathymetric Synthetic Aperture Sonar

Test Results from a Multi-Frequency Bathymetric Synthetic Aperture Sonar M. P. Hayes, P. J. Barclay, P. T. Gough, and H. J. Callow Acoustics Research Group Department of Electrical and Electronic Engineering, University of Canterbury, Christchurch, New Zealand. Abstract- This paper describes the implementation of a bathymetric synthetic aperture sonar and presents preliminary results from sea trials of the sonar. The sonar is designed for high resolution seafloor imaging in a shallow water environment. This is achieved through coherent summation of successive echo signals to synthesise an aperture many times longer than the towfish. Provided the motion of the towfish is accurately estimated and compensated, the application of aperture synthesis can result in a range independent resolution over the operating swath. While high resolution is desirable for high quality imagery, it is not always sufficient for the discrimination of targets of interest from general seafloor clutter. Therefore, we have attempted to use bathymetric techniques to discriminate targets from the seafloor on the basis of height. The sonar is configured to use an array of three vertically spaced hydrophones from which three independent synthetic aperture images are reconstructed. Using knowledge of the sonar geometry, the relative heights of the seafloor scatterers are estimated using the phase differences resulting from the slight range difference of a scatterer from each hydrophone. Preliminary results are presented from trials in a shallow water (nominally 16 m) harbour environments. Results are shown for two different frequency bands: the sonar transmits two simultaneous linear FM chirp signals; one covering the frequency band of - khz and the other covering 9-11 khz. Due to the use of a neutrally-buoyant, high-drag towfish, reasonable reconstructed imagery is obtained without autofocusing at the low frequency range. However, autofocusing is required for the high frequency data even though the synthetic apertures are shorter. I. INTRODUCTION The earliest sonars were no more than primitive, hullmounted depth-sounders that measured the time delay between a transmitted pulse of sound (the ping) and pulse echoes returned from reflection off the seafloor. These time delays, when scaled by the speed of sound, gave an estimate of the depth beneath the vessel. Since a single pulse-echo was of little value, the sonar sent a ping, waited for a suitable length of time to record all the returns from the seafloor and then repeated the ping transmission (usually endlessly) to get very precise measurements of the depth at that point. Recording spot depths at a specific position on a chart became the modern science of bathymetry. Soon this simple system was extended to have a single transmitted beam and a cross-hull array of hydrophones. With this arrangement, a fan of beams could be synthesised to produce a multi-beam echosounder. Now with each pass of the surface vessel could plot out 1 or 12 slant depth estimates from the vessel to different tracks across the seafloor (the true depth referred to some x,y coordinate needed some extra slant angle correction). However what was of great value was that, in a single pass of the vessel, a crude, low-resolution image of the seafloor could be obtained as well as precise depth estimates over that image. In this one system, there was a coming together of bathymetry with seafloor imagery. In a parallel development, the simple one-beam, echosounder was mounted on a cable-towed fish and towed along underneath the sea surface. However, now the sonar was set to radiate and receive in the horizontal plane normal to the direction of travel and so was called a side-looking or sidescan sonar. As the surface vessel traversed the track, the sonar beam swept out across the terrain producing a range versus reflectivity trace for each ping. Recording this trace side-by-side on a line printer as the vessel traversed a straight-line track at a constant speed gave a clear impression of the seafloor features. As an image of the seafloor, it had certain flaws. Although the range resolution (the resolution along the beam) was acceptable, the azimuth resolution (across the beam) degraded with increasing range. This is unavoidable in a single-beam system since there is beam spreading in azimuth. Despite the problem of rangedependent azimuth resolution, the images produced were reasonably clear and have formed the basis of many seafloor surveys. Recently, mostly based on the successful deployment of Synthetic Aperture Radar (SAR) systems, side-looking sonars have been modified to record both the amplitude and the phase of the returned pulse echoes. With this added feature, several echo returns could be added together coherently in order to synthesise an apparently larger aperture than the real aperture on the sonar towfish. Not surprisingly this is now called a Synthetic Aperture Sonar (SAS). The elegance of this approach is that the number on pulse-echo returns added could be increased with range, the along-track resolution could be held constant. Moreover, this happened automatically since closer reflecting targets were insonified for fewer number of pings than targets at further ranges. This meant that given a block of pulse-echo returns for many pings, they could be processed so the along-track resolution remained constant at all ranges. In the limit the cross-track MTS -933957-28-9 1

resolution was given by c 2 B where c is the speed of sound, B is the bandwidth and the along-track resolution was given by half the extent of the real aperture. With suitable choice of parameters, the cross and along-track resolution could be made equal and produce more optical like quality images of the seafloor. Of course there were complications in this apparently ideal solution of side-looking sonar imagery based on SAS. The major complication with SAS is that the towfish has to travel along a very straight track. This is necessary so the coherent (i.e., phase sensitive) addition of multiple pulseecho returns add up in the correct fashion. A second complication is that the vessel can move not much further than about the along-track resolution between pings [1]. Consequently, the more precise the resolution required, the slower the vessel is constrained to travel. In a similar way to the development of the multi-beam echo-sounder, the solution to the second complication is to mount an array of many hydrophones along the towfish in the direction of travel. Now the sonar system can move at a more useful velocity; approximately the speed of the single hydrophone system multiplied by the number of hydrophones in the array. This single transmitter, multiple hydrophone linear array is the basis of most SAS in operation today. But there is still something missing. What is still to do is to produce an estimate of the depth of the seafloor as well as the imagery. The equivalent has been done with SAR using interferometry based on a two element array in the height direction [2]. Two images are created each from slightly different heights and by cross correlating the two images a crude estimate of the terrain profile can be realised. This approach has yet to be applied to a free-towed underwater SAS although some preliminary attempts have been made with rail-based SAS [3]. This coming together of crude bathymetry with precise SAS imagery would then truly complement the multi-beam echosounder's ability to produce crude imagery with precise bathymetry. In this paper, we describe the KiwiSAS-III synthetic aperture sonar, present methods used to compensate for the towfish motion, and then show results obtained from sea trials with an unconstrained tow. Finally, conclusions are presented with suggestions for future research. II. KIWISAS-III SONAR The KiwiSAS-III synthetic aperture sonar uses a similar towfish to its predecessor, KiwiSAS-II [4], but with multiple receiver hydrophones, modified sonar electronics, and a custom inertial navigation system (INS). A. Towfish The shell of the sonar towfish is a PVC cylinder, of length 13 mm and diameter 249 mm, with an aluminium keel and a total weight of 52 kg in air. It is of a neutral buoyancy design with a sacrificial depressor chain attached to the sonar using galvanic releases so that the sonar can be Fig. 1. The synthetic aperture sonar towfish. The hydrophone is comprised of an array of three vertically displaced elements. recovered in the event of a tow cable failure. The towfish has a blunt nose and is towed from the nose to minimise yaw. As illustrated in Fig. 1 the hydrophone array is mounted in a keel and separated from the projector to reduce acoustic crosstalk. The hydrophone array consists of nine 75 mm square PVDF tiles connected as three rows of three tiles, giving three vertically displaced 215 mm by 75 mm hydrophones. Each hydrophone has its own preamplifier and cable driver. The projector transducer is 343 mm long and 9 mm high. It consists of an array of 3 12 Tonpilz transducers with the most efficient transducers positioned in the centre of the array. This provides some aperture shading to reduce the sidelobe levels of the transmitted beam pattern. The transducers were designed to be resonant between - khz but were also found to operate well between 9-11 khz. The power amplifiers for each of the transducer elements are mounted within a stainless steel cylinder within the towfish and operate off a 24 V supply from the towboat and supplemented with a capacitor bank in the towfish. The towfish is instrumented with a three axis magnetometer and accelerometer navigation system, supplemented with a two-axis clinometer and pressure transducer. On board microcontrollers convert the measurements into NMEA format messages that are transmitted to the towboat and logged along with GPS messages recording the towboat position. B. Sonar electronics The heart of the sonar electronics consists of an array of digital down converters that are time division multiplexed and interfaced to an off-the-shelf personal computer using a custom high-speed PCI serial interface card. Each receiver channel has a pair of digital down converters tuned to the two different frequency bands. The gain of each channel can be digitally controlled. The transmitted signal is produced by an arbitrary waveform generator, essentially a large field memory connected to a digital to analogue converter. The transmitted signals are stored on disk in their baseband representations and then modulated onto the desired carrier frequencies. The 2

carrier frequencies are selected so that continuous sampling can be used with no dead-time between transmitted pulses. Thus aliased targets outside the maximum unambiguous range can be processed. III. MOTION COMPENSATION With a free-towed sonar, motion compensation of the data is usually required in all but the calmest harbour environments. In particular, timing errors need to be corrected to ensure phase coherence across the synthetic aperture being reconstructed. These timing errors are primarily due to the sway motion of the towfish when operating in shallow waters. For bathymetric measurements using phase monopulse techniques, the critical towfish motion that must be corrected is roll. Motion compensation of the data is performed using three independent correction steps: 1. Bulk yaw correction. 2. Sway correction. 3. Roll correction. While the motions of the towfish are coupled, it is simpler to assume that they are independent. The bulk yaw correction is required before the sway correction step otherwise any constant yaw of the towfish will be estimated as a progressively increasing sway. The approach we use is to calculate the along-track spatial frequency response of the echo data, averaged across the range bins. We then correlate this estimated spatial frequency response with the expected spatial frequency response using knowledge of the beam patterns. The displacement of the correlation peak provides an estimate of the spatial frequency offset resulting from towfish yaw. Note that the correlation approach is preferable to calculating the centroid of the spatial frequency response[5], especially when the synthetic aperture is undersampled. The sway is then estimated using a weighted shearaverage[6], with the weighting selected to suppress both system noise and strong targets. The autofocus techniques we use in this paper are presented in more detail by Callow[7,8]. For roll correction, we calculate the phase differences between the top and middle, middle and bottom, and top and bottom transducers, with a phase correction applied for the expected slant range geometry using estimates of the towfish depth and the towfish height above the seafloor[9]. A roll angle estimate is then formed from a weighted least means squares estimate of the phase differences for each range bin. IV. RESULTS The experimental data presented in this paper were collected in the Hauraki Gulf, near Auckland, New Zealand during the first few days of June 1 (start of Winter). The conditions were sea state 3 with a 1.5 m confused chop and knot winds. The sonar was towed behind a small New Zealand Navy work boat at speeds in the range of 1.5-4.5 knots. The data was collected in the middle of the day for about an hour either side of high tide where the water depth was approximately 16 m. The seafloor is predominantly mud and essentially flat. The operating depth of the towfish depends on the tow speed, the length of cable, and the weight of the depressor chain. Typically, it was operated in midwater to give the best seafloor echo coverage. The sonar transmitted two simultaneous linear FM chirps: one in the frequency range of - khz and the other in the frequency range of 9-11 khz. The chirps were either 12.5 ms in duration, repeated every 68.8 ms giving an unambiguous range of 51.6 m or 25. ms in duration repeated every 34.4 ms doubling the unambiguous range to 13.2 m. Each of the received echo signals from the three hydrophones were demodulated to baseband using a pair of digital down converters per hydrophone; with one down converter tuned to select the low frequency chirp and the other tuned to select the high frequency chirp. The six output channels were then logged to disk in real-time. Fig. 2 shows magnitude of the pulse compressed echo data obtained from the top hydrophone for a sequence of 512 pings over a second interval, covering an along-track distance of approximately 75 m. The image gray scale, of this and the other images presented in this paper, is linearly related to the image magnitude but with clipping at 1% of the peak pixel magnitude. The vertical axis represents across-track distance in metres and the horizontal axis represents along-track distance in metres. The section of data displayed was chosen from many hours of recorded data due to the presence of the interesting target at a range of about m. There is obvious along-track smearing of the target due to the wide beam widths of the sonar at khz. Predictably, the companion 1 khz image, shown in Fig. 3, shows less smearing due to the narrower beamwidths. The reconstructed synthetic aperture images, without autofocus correction, from the khz and 1 khz pulse compressed data are shown in Fig. 4 and Fig. 5. Surprisingly, given the sea conditions, an improvement in along-track resolution can be seen with the khz data. However, with the 1 khz image the reconstructed image is poorer than the raw image. Also note the shift in target position due to the constant towfish yaw. For the results presented here, the towfish speed was estimated from a boat GPS log and from global contrast optimisation techniques [1,11] to be travelling at approximately 2.1 m/s. Even with a fast pulse repetition rate (and a consequently short maximum unambiguous range), this speed translates to an along-track sampling interval of 2D 3 where D is the along-track dimension of the hydrophones. Ideally, for high quality synthetic aperture imagery, the along-track sampling interval should be smaller 3

Fig. 2. Magnitude of pulse compressed echo data from the top hydrophone at khz. The vertical axis is cross-track range and the horizontal axis is along-track position, both in metres. The image gray scale is linear, clipped at 1% of the maximum pixel value. Fig. 4. Magnitude of pulse compressed echo data from the top hydrophone at khz reconstructed using the wavenumber algorithm. Fig. 3. Magnitude of pulse compressed echo data from the top hydrophone at 1 khz. than D 3 and preferably smaller than D 4 to avoid image artefacts due to grating lobes in the synthetic beam pattern[1]. To improve the quality of the reconstructed images, autofocusing was applied to estimate the towfish sway. First, however, the bulk yaw was estimated using a weighted least means square average from the three hydrophones using the khz data. This was found to be 5 degrees forward; approximately half the null-to-null beamwidth of the sonar at khz. This constant yaw was then removed from the image data for all the hydrophone channels and then an estimate of the towfish sway was found using a weighted least means square average of the sways estimated from each image. Both the 1 khz and khz data was used, with the khz data used to disambiguate 2π phase jumps in the 1 khz data. The sway that was estimated is plotted in Fig. 6. Note Fig. 5. Magnitude of pulse compressed echo data from the top hydrophone at 1 khz reconstructed using the wavenumber algorithm. that due to the aperture undersampling and large sway errors, there were regions along the aperture where it was necessary to unwrap the phase differences resulting from the shear averages. One thing to note from the sway estimate plotted in Fig. 6 is that there is significant curvature in the path of the towfish but little high frequency content in the region of the target of interest. After correcting for the sway motion, by applying a linear phase with frequency in the temporal frequency domain, and then re-applying the wavenumber algorithm to reconstruct a synthetic aperture image, the results shown in Fig. 7 ( khz) and Fig. 8 (1 khz) were obtained. With the khz image, there is a slight improvement in alongtrack resolution of the targets of interest but a more marked contrast sharpening in other parts of the image. There is a more significant improvement with the autofocussed 1 khz image. 4

.3.2.1 Sway Estimate (m).1.2.3.4.5.6 1 Fig. 6. Estimated towfish along-track sway using a weighted shear average of the khz data. Fig. 8. Magnitude of pulse compressed echo data from the top hydrophone at 1 khz reconstructed using the wavenumber algorithm, with bulk yaw correction and sway correction. Pitch (deg) 5 15 Roll (deg) 1 4 6 8 12 1 7 Fig. 7. Magnitude of pulse compressed echo data from the top hydrophone at khz reconstructed using the wavenumber algorithm, with bulk yaw correction and sway correction. A preliminary attempt has been made to generate a bathymetric image of the seafloor from the measured data. The first step was to create interferograms between each of the images from the displaced hydrophones. For example, Fig. 11 shows the phase difference between the top and and middle hydrophones. The most obvious feature is the banding due to a rolling motion of the towfish. This roll induced phase error swamps any phase differences due to target height and thus it was necessary to estimate the roll and to correct the interferograms. Depth (m) 8 9 11 1 Fig. 9. Towfish pitch and roll measured using clinometers and towfish depth estimated from pressure measurements. The roll of the towfish was estimated by comparing the phase difference between the vertically displaced hydrophones and plotted against the measurements from a clinometer within the towfish. As can be seen from Fig. 1 there is very good agreement between the clinometer and the roll estimated from the data phase using both the khz and 1 khz echo signals (the standard deviation of the roll difference is.6 degrees at 1 khz and.5 degrees at khz). With the 1 khz data there was an angular ambiguity that was resolved using the khz data. The roll corrected interferogram is shown in Fig. 12. The banding can be seen to have been removed although there is still a slant range dependent phase gradient. Unfortunately, the targets of interest had only a slight phase difference from that of the surrounding flat seafloor, suggesting that the targets of interest were only slightly proud of the seafloor. 5

2 5 1 4 6 Roll Estimate (degrees) 15 8 25 a. INS roll estimate b. khz data roll estimate c. 1kHz data roll estimate 12 14 1 Fig. 1. Towfish roll estimated from both the khz and 1 khz data superimposed on the roll measured by a clinometer within the towfish. 1 Fig. 12. Phase difference image formed from top and middle hydrophones at khz after roll correction. 5 While we have shown an improvement in along-track resolution, the image quality from this trial is poorer than what we have achieved from other trials. This is likely to be attributable to the sea conditions at the time (see Fig. 9 for a plot of the measured pitch and roll of the towfish). In particular, we had to travel faster than what we intended so that we could maintain headway in the strong winds with a small boat. This lead to an undersampled aperture degrading the reconstructed image quality. This is compounded by a digital hardware fault in the sonar that was retrospectively discovered to have introduced unwanted interference on some of the channels, primarily the khz channels. 1 15 25 Currently, we are working on reconstructing seafloor and target height estimates from the differential phase data from the hydrophone array. Further work is required for modelling the towfish motion and fusing the motion measured by the navigation unit with the motion estimated using the image data. This could be used to refine the motion estimated from the image data. 1 Fig. 11. Phase difference image formed from top and middle hydrophones at khz showing effect of roll. ACKNOWLEDGMENTS V. CONCLUSIONS We would like to acknowledge the assistance of the New Zealand Defence Technology Agency and the New Zealand Navy for their assistance with the sea trials in the Hauraki Gulf that provided the experimental data for this paper. In this paper we have presented experimental results from a free-towed dual-frequency synthetic aperture sonar. The sonar has three vertically displaced hydrophones that can be used in a phase monopulse configuration to estimate target heights. However, since there are no additional horizontally spaced hydrophones, the sonar has a limited unambiguous range, constrained by the slow speed of sound. We would like to thank Edward Pilbrow and Alan Hunter for their assistance with building and analysing the navigation unit, Peter Lambert for constructing the towfish, Michael Cusdin for assisting with the towfish electronics, and Steven Fortune for his efforts during the sea trials under trying circumstances. Even when using an unconstrained tow in unfavourable sea conditions, there is sufficient information from the seafloor clutter to estimate both sway and roll motions of the sonar to an accuracy required to improve the imagery. The use of dual frequency bands has been advantageous for the towfish motion estimation since the low frequency data helped disambiguate the phase of the high frequency data. Philip Barclay is supported by a University of Canterbury Doctoral Scholarship; Hayden Callow and Steven Fortune are supported by Bright Futures Top Achiever Doctoral Scholarships from the New Zealand Foundation of Research, Science, and Technology. 6

REFERENCES [1] D. W. Hawkins and P. T. Gough, Imaging algorithms for a strip-map synthetic aperture sonar: minimizing the effects of aperture errors and aperture undersampling, IEEE Journal of Oceanic Engineering, vol. 22, pp. 27 39, Jan. 1997. [2] C. V. Jakowatz, Jnr., D. E. Wahl, P. H. Eichel, D. C. Ghiglia, and P. A. Thompson, Spotlight-Mode Synthetic Aperture Radar: A Signal Processing Approach. Boston: Kluwer Academic Publishers, 1996. [3] Y. Perrot, B. Hamonic, and M. Legris, Three-dimensional high resolution imaging using sas and interferometric process, in Fifth European Conference on Underwater Acoustics, vol. 1, pp. 425 431, July. [4] D. W. Hawkins and P. T. Gough, Recent sea trials of a synthetic aperture sonar, Proceedings of the Institute of Acoustics, vol. 17, pp. 1 1, Dec. 1995. [5] F. Berizzi, G. Corsini, M. Diani, and M. Veltroni, New time-domain clutter-lock algorithm, in IEE Proceedings on Radar, Sonar, and Navigation, vol. 144, pp. 341 347, IEE, December 1997. [6] K. A. Johnson, M. P. Hayes, and P. T. Gough, A method for estimating the sub-wavelength sway of a sonar towfish, IEEE Journal of Oceanic Engineering, vol., Oct. 1995. [7] H. J. Callow, P. T. Gough, and M. P. Hayes, Autofocus of multi-band, shallow-water, synthetic aperture sonar imagery using shearaveraging, in IGARS1, International Geoscience and Remote Sensing Symposium,(Sydney, Australia), July 1. [8] H. J. Callow, P. T. Gough, and M. P. Hayes, Noncoherent autofocus of single-receiver broad-band synthetic aperture sonar imagery in OCEANS1,(Honolulu, Hawaii), Nov. 1. [9] M. P. Hayes and T. Y. Ho, Height estimation of a sonar towfish from sidescan imagery, in IVCNZ, (Hamilton, New Zealand), pp. 12 17, Nov.. [1] S. A. Fortune, P. T. Gough, and M. P. Hayes, Statistical autofocus of synthetic aperture sonar images using image contrast optimisation, in IGARSS1, International Geoscience and Remote Sensing Symposium, (Sydney, Australia), July 1. [11] S. A. Fortune, P. T. Gough, and M. P. Hayes, Statistical autofocus of bathymetric, multiple-frequency synthetic aperture sonar images using image contrast optimisation, in OCEANS1, (Honolulu, Hawaii), Nov. 1. 7