Detection and Classification of Underwater Targets by Echolocating Dolphins. Whitlow W. L. Au

Transcription

1 Detection and Classification of Underwater Targets by Echolocating Dolphins Whitlow W. L. Au Hawaii Institute of Marine Biology University of Hawaii Abstract Many experiments have been performed with echolocating dolphins to determine their target detection and discrimination capabilities. Target detection experiments have been performed in a naturally noisy environment, with masking noise and with both phantom echoes and masking noise, and in reverberation. The echo energy to rms noise spectral density for the Atlantic bottlenose dolphin (Tursiops truncatus) at the 75% correct response threshold is approximately 7.5 db. The dolphin s detection threshold in reverberation is approximately 2.5 db. Echolocating dolphins also have the capability to make fine discriminate of target properties such as wall thickness difference of water-filled cylinders and material differences in metallic plates. The high resolution property of the animal s echolocation signals and the high dynamic range of its auditory system are important factors in their outstanding discrimination capabilities. Time differences between echo highlights at small as ns can be resolved by echolocating dolphins. They also will attend to echo components that are db below the maximum level for a specific target. These capabilities coupled with their capability to adaptively vary their search trajectory depending on echoes just previously received may also play a big factor in their detection and discrimination capabilities. 1. Introduction Experiments with echolocating dolphin have shown that these animals have remarkable sonar capabilities when compared with technological sonar [1]. Unlike any man-made sonar, dolphins have a unique capability to discriminate fine differences in target features which allow them to recognize specific targets, even targets that are buried in ocean sediment. This target recognition capability seems to surpass those of the most advance technological sonar systems. The properties of the dolphin sonar system will be reviewed and their sonar detection and discrimination capabilities will be discussed. From an acoustic perspective the properties of the dolphin sonar system are not outstanding and could even be characterized as being rather unimpressive. However, dolphins can perform sonar tasks that far outperform technological sonar systems. In this paper, an attempt will be made to explain why dolphins can perform sonar tasks that we might consider to be rather remarkable. The echolocation system of a dolphin can be divided into three major subsystems, the reception, transmission, and signal processing/decision making subsystems. The receiving subsystem consists of the auditory system of the animal, and its capabilities depend on the characteristics of the peripheral and higher auditory centers of the auditory central nervous system. The capability of a dolphin to detect objects in noise and clutter and to discriminate between various objects, and to recognize specific objects depends to a large extent on the information-carrying capabilities of the emitted signals. Also important are the extent to which the dolphin s auditory system can extract pertinent information from the echoes and the animal s cognitive capabilities. In order to make optimal use of acoustical information, the dolphin should have an auditory system that is very sensitive over a wide frequency range. The dolphin should also be sensitive in both quiet and noisy environments and should be able to detect short- and long duration sounds. A good spectral analysis capability is important in discriminating and recognizing predators, prey, and other objects in the environment. Other important characteristics of a good sonar receiver include the ability to spatially resolve and localize sounds, reject externally generated interferences, and recognize temporal and spectral patters of sounds. Most of the data that will be discussed here come from the Atlantic bottlenose dolphin, Tursiops truncatus. This species is the most common in oceanariums, marine parks, aquaria and other public display facilities. It is also the species that is most common in captivity. Occasionally, data from other species will be used when appropriate. 2. Receiver Characteristics A number of hypotheses have been proposed on how sounds propagate from the water into the inner ears of dolphins. The most prevelant theory was proposed by Norris [2],[3] in which sounds entered into the oval window on both sides of the lower jaw and propagate along fat channels into the inner ears. In order to examine the hypothesis of Norris, Møhl et al. [4] perform an experiment in which a piezoelectric ceramic embed- We3.A III

2 ded into suction cup was placed at different location about a dolphin s lower jaw and the threshold of hearing was determined. The results of Møhl et al. [4] are shown in Fig. 1. The area of high sensitivity was found to be forward of the oval window. From a transduction perspective, the lower jaw of the dolphin can be considered as a shaded receiver or hydrophone. The auditory system of the dolphin does not receive sounds propagating from different directions equally well, but instead receive sounds in a beam [7]. The receiving beam pattern of a bottlenose dolphin was measured by Au et al. [7] and their results in both the vertical and horizontal planes for three different frequencies, 30, 60 and 120 khz are shown in Fig. 3. The major axis in the vertical plane is pointed between 5 and 10 o above the horizontal axis. In the horizontal plane, the beam axis is pointed directly in front of the dolphin. Figure 1. Points of stimulation with attenuation values in db required for a threshold. The larger the atteuation value, the more sensitive the area. The area marked by the dashed line is the oval window of the lower jaw. 2.1 Hearing sensitivity The hearing sensitivity at different frequencies (audiogram) of a bottlenose dolphin was measured in a classic study by Johnson [5]. His results along with those of Au et al. [6] are shown in Fig. 2. The audiogram of the SOUND PRESSURE LEVEL (db re 1 µpa) Johnson[5]) 120 Au et al. [6] FREQUENCY (KHZ) Figure 2. Hearing sensitivity of two Atlantic bottlenose dolphins as a function of frequency. of the two dolphins indicate that dolphins have a very broad frequency range of hearing from 100 Hz to 150 khz, covering approximately 10 octaves. The maximum sensitivity is approximately 40 db re 1 µpa, which close to sea state 0 when taking into consideration the filter bandwidth in the dolphin s auditory system. 2.2 Receiving beam pattern 200 Figure 3. Receiving beam pattern for the bottlenose dolphin for three different frequencies (from Au and Moore[7]). Au and Moore [7] using the beam patterns of Fig. 3 calculated the directivity index at 30, 60 and 120 khz and found that the following line fitted the results well DI = 16.9 Log f 14.5 (2) Where f is in khz and DI is in db. For a frequency of 120 khz the directivity index is equal to 20.5 db. 2.3 Auditory filter shape The auditory filter shape of a mammalian subject can be determined by performing a notched noise masking experiment where the tone signal is directly in the middle of the notch. A schematic of a notched masking noise along with a tonal signal is shown in Fig. 4. The variable g in Fig.4 is the frequency ratio g f f f o = (3) where f o is the frequency of the tonal signal. Only the noise in the shaded portion that is under the W(g) curve will contribute to the masking of the tonal signal. Therefore, at the threshold of hearing of the tonal signal the acoustic power in the tonal signal (P s ) is equal to the III

3 NOISE g W(g) TONE NOISE propagates into the water. The melon also has an impedance matching function for signals generated within the nasal area and propagate through the fatty melon. 3.1 Signal waveform and spectrum -g 0 g Figure 4. Notched masking noise along with a tonal signal and an arbitrary filter function W(g). power in the masking noise that are under the W(g) curve so that g P = KN W ( g) dg s 2 (4) o Bottlenose dolphins emit short broadband clicks having peak frequencies as high as khz [1]. Signals duration vary from 40 to 70 µs, having 4 to 10 positive excursions. Peak-to-peak source levels between 210 and 227 db re 1 µpa have been measured [1]. Bandwidth can vary between 40 and 60 khz. Dolphins in tank naturally emit much lower level signals with lower peak frequency. Examples of echolocation signals bottlenose dolphins are shown in Fig. 5. where K is a constant and N o is the noise spectral density. Equation 4 can be differentiated to give 1 dps W ( g) = (5) 2KN dg Therefore, if the threshold of hearing about a specific tonal frequency is determined as a function of the width g of the notched noise, Eq. 5 can be used to determine the auditory filter shape. Such a study was performed by Lemonds et al. [8] and their results are shown in Fig. 5 for frequencies of 40, 60, 80 and 100 khz. o AMPLITUDE (db) FREQUENCY (khz) Figure 5. Auditory filter shape for a bottlenose dolphin. Note that the filters are not very narrow. The shapes are similar to that of humans if we normalized the frequency by dividing by f o for any given filter. If the 3- db bandwidth is plotted as a function for each filter then a Q of 8.4 can be fitted through the bandwidth points. 2. Tranmitter Characteristics Echolocation signals are produced within the nasal system of dolphin and the sounds propagate through the melon into the water [9]. The melon of the dolphin contain a special lipid material that is found there and behind the oval-window and extends directly to the auditory bulla [9]. This type of fat has been referred to as acoustic fat. The melon has a layered structure with a a low-density core surrounded by progressively denser fat. The structure of the melon has been hypothesized to focus acoustic energy within the melon as the signal Figure 5. Representative echolocation signals of Tursiops truncates in a tank and in open waters. The waveforms are on the left and the frequency spectra on the right (from Au [1]). 3.2 Transmitting beam pattern Signal are transmitted in a beam as shown in Fig. 6. The waveform of the signal measured by hydrophones at different angles about the animal s head is also shown. The transmit beamwidth of a high frequency sonar signal is 10.2 o (horizontal plane) and 9.7 o (vertical plane). The receiving beamwidth is slightly wider, 13.7 o and 17 o in the horizontal and vertical planes, respectively. The directional projection and reception characteristics of bottlenose dolphin are not exceptional compared to many technological sonar One of the property of broadband signals is the distortion of off-axis signals as can been seen in Fig. 5 for both planes. When a signal is measured at an angle greater than about 5 o away from the beam axis, the signals become distorted and the amount of distortion increases as the angle increases. The transmit beam patterns are slightly narrower than the receive beam patterns for a frequency of 120 khz. III

4 solid steel sphere [11] and a 7.62-cm diameter waterfilled sphere [12] 73 m and 113 m, respectively. The target detection capability of the bottlenose dolphin was measured by two other techniques. A target was positioned at a fixed range and the dolphin s ability to detect it was measured as a function of the level of a wide-band making noise [1]. In another experiment an electronic simulated echo generator was used to simulate a phantom target at 20 m and the level of the echo was progressively made smaller as the dolphin performed a detection task. The results of the three different types of target detection experiments are displayed in Fig. 7 showing correct response versus the echo energy-to-noise ratio. At the 75% correct response threshold the echo energy-to-noise is approximately 7.5 db. Detection threshold based on the 75% correct response level includes both correct detection and correct rejection trials and is roughly equivalent to the 50% correct detection threshold. Figure 6 Transmission beam pattern for a bottlenose dolphin in the vertical and horizontal planes (from Au [1]). 4 System s Performance The capabilities of any echolocation system are usually divided into two general categories; target detection and target discrimination. The target detection capabilities of an echolocating dolphin will be limited by ambient noise or artificial noise, reverberation, and by its own hearing sensitivity. Echolocation experiments to determine the target detection capabilities of dolphins in noise and reverberation have been performed by Au [1]. Many different types of discrimination experiments have been performed (see Nachtigall [9] and Au [1]). We will only discussed a few of these experiments, two which may provide us with insight into cues that dolphins used to discriminate targets. 4.1 Target detection in noise Three different types of target detection in noise experiments have been performed with dolphins. The first type is the simple one in which a specific target is moved progressively farther away from the dolphin until the animal can no longer detect the target. The maximum detection range of two bottlenose dolphin measured in Kaneohe Bay using a 2.54-cm diameter Figure 7. Bottlenose dolphin performance results with correct response plotted against the echo energy to noise ratio. The solid line is the energy detection model of Urkowitz [13]. In order to obtain a better appreciation of the detection process of the bottlenose dolphin, consider the echo in noise schematic of Fig. 8. The largest echo highlight is observable in the noisy echo, however the smaller high lights are masked by the noise and the acoustic quality of the echo is altered. The dolphin could probably hear the largest highlight but the echo probably did not sound like the sphere they were trained to detect and consequently reported the target as not present. Therefore, it seems that a target detection experiment is probably not purely one of detecting signal in noise but also involves discriminating the features of the echoes from a target. III

5 the highest highlight of the target echo is clearly detectable, however the lower highlights are masked by the reverberation. The dolphins could probably hear the larger highlights from the target but the echo probably did not sound like the target they were trained to detect and consequently they reported the target as not present. As in the noise case, target detection in reverberation also involves recognition of the target echo. 4.3 Discriminating composition and thickness of metallic plates Figure 8. Target echo in noise at the dolphin s threshold. The sphere echo is shown in the top trace, and the echo mixed with noise in the 2 nd and 3 rd traces. The noise in the 2 nd trace is not filtered while the noise in the 3 rd trace is filtered between 75 and 150 khz. 4.2 Target detection in reverberation A second way in which a sonar system can be limited is by the presence of reverberation. Reverberation differs from noise in several aspects. It is caused by the sonar itself and is the total contribution of unwanted echoes scattered back from objects and inhomogeneities in the medium. Murchison [11] studied the effects of bottom reverberation on the target detection capabilities of two bottlenose dolphin in Kaneohe Bay. A 6.35-cm diameter solid steel sphere was used and placed on the bottom. The animals 50% correct detection threshold range was approximately 70 m. Au [13] used a simulated dolphin sonar signal to measure the scattering strength of the bottom where Murchison [11] performed his experiment. Taking the target strength into consideration and the difference in the transmit and receive beam patterns of the transducer and the dolphin the reverberation form of the sonar equation was used to estimate an echo energy-to-reverberation (E/R) of approximately 4 db. The measurements of Au [13] are shown in Fig. 13. Again as in the masking noise case, Figure 9. Target echo in reverberation at the dolphin s threshold of detection (from Au [13]). Evans and Powell [14] demonstrated that a blindfolded, echolocating bottlenose dolphin could discriminate between metallic plates of different thickness and material composition. The dolphin was trained to recognize a 30-cm diameter circular copper disc of 0.22-cm thickness from comparison targets of the same diameter A schematic of the dolphin performing a typical search and the various comparison material and plate thickness are shown in Fig. 10. The dolphins could perform the task well above chance. Figure 10. The left panel shows a typical sonar search by the blindfolded bottlenose dolphin and the right side of the figure shows the various comparison targets comparison target used by Evans and Powell [14]. Au and Martin [15] examined the plates used in the experiment of Evans and Powell [14] with an echo ranging system that projected simulated dolphin echolocation signals. Backscatter results at normal incident indicated virtually no cues for discrimination was present in the echoes. However, when the plates were examined at angles away from the normal, the different plates began to display unique highlight structures. Examples of backscatter at normal incident and at 14 o incident are shown in Fig. 11. The echoes from the 14 o incident angle are about 20 db below that of the normal incident, yet the discrimination cues were present for the off-axis backscatter. This implies that dolphin are able to use cues that are at least 20 db below the maximum amplitude of of the echoes at normal incident in order to discriminate targets. III

6 The various discrimination experiments with echolocating dolphins strongly suggest that these animals possess a sophisticated and well honed sonar system. However, when some of characteristics of the different compo- θ = 14 o The echo waveform, envelope, and frequency spectrum for the standard and the comparison targets having a wall thickness difference of 0.3 mm are displayed in Fig. 13. The dolphin was able to perform above threshold for the targets represented in Fig. 13. The envelope curves suggest that if the dolphin used time domain cues, it may be able to perceive incremental time differences of approximately 0.5 µs between highlight intervals. Differences in the frequency spectra of the echoes can also be seen in Fig. 13. The spectrum for the thinner comparison target resembled the spectrum of the standard target but was shifted slightly toward lower frequencies. The spectrum for the thicker comparison targets were shifted toward higher frequencies. The Figure 11. Examples of backscatter from the standard disk and some of the comparison disks used by Evans and Powell [14]. 4.4 Cylinder wall thickness discrimination The capability of a bottlenose dolphin to discriminate the wall thickness differences was measured by Au and Pawloski [16]. A dolphin was trained to station in a hoop and echolocate of two targets 8 m away separated by 22 o azimuth. The standard target was a 3.81-cm O.D. aluminum cylinder with a wall thickness of 6.35 mm. Comparison targets with wall thickness both thinner and thicker than the standard were used. The comparison targets had incremental differences in wall thickness of ± 0.2, ± 0.3, ± 0.4 and ± 0.8 mm from the standard target. The dolphin was required to echolocate and to respond to the paddle that was on the same side of the center line as the standard target. The dolphin s performance as a function of wall thickness difference is shown in Fig. 12. The 75% correct response threshold corresponded to a wall thickness difference of 0.23 mm for the thinner targets and mm for the thicker targets. Figure 13. Echo waveform, waveform envelope, and frequency spectrum for the standard and comparison target having a wall thickness difference of 0.3 mm. The dashed envelope and spectrum curves are for the comparison target(from Au and Pawloski [16]). average frequency differences were 3.2 and 2.2 khz for a wall thickness difference of 0.3 and 0.2 mm, respectively. If the dolphin used this shift in frequency spectra to discriminate the wall thickness difference, then the spectral data suggest that the dolphin could perceive a shift of approximately 3.3 khz, but not a shift of 2.1 khz. 5. Discussion and Conclusions Figure 12. Dolphin wall thickness discrimination performance (from Au and Pawloski, [16]). III

7 nents are examined, it is surprising that dolphins can echolocate so well. The transmit and receive beam patterns are not very narrow. The auditory filters are also not very narrow. There are many technological sonar with both narrower beams and narrow filters. So we are left with the question on what are the factors that allow the dolphin sonar to be have such good discrimination and recognition capabilities. The use of broadband echolocation signals certainly contributes to the dolphin s discrimination capabilities. The temporal resolution of a dolphin echolocation signal is approximately µs, which translate to a distance resolution of about m or 15 cm. However, the short duration of the transmit signal limits the amount of energy within a signal so that the range of the sonar system is not very large. Typically, dolphins seem to be interested in object that are within 100 m and are hardly concerned about longer ranges. So in a sense, temporal resolution was traded off with maximum range in the evolution process. However, within a 100 m range, there is not a technological sonar that can rival the dolphin in discriminating and recognizing targets. Bottlenose dolphins can even detect and discriminate target that are buried in ocean sediment. The use of broadband signals may also allow for the perception of time-separation pitch by dolphins. When a sound consisting of two correlated pulses are projected to humans, a pitch that is equal to the reciprocal of the time delay between the two pulses can be perceived by the human auditory system and may also be perceived by most mammal. If more than two highlights are present in an echo, a time-separation like pitch can still be heard. Therefore, a dolphin may discriminate targets from a pitch-like sound that multi-highlight echoes produce. A second feature of the dolphin sonar system that is used to great advantage by the animals has to do with the dynamic range of its system. The metallic plate discrimination experiment discussed in section 4.3 suggest that dolphins may gain information on an object by examining echoes that are 20 to 30 db below the maximum level associated with a particular target. In other words, a dolphin does not seem to only seek out specific orientations to a target that will produce the highest echo levels but orientations that will provide the most information. In the specific example of section 4.3, the orientation to the target that provided valuable information came from incident angles that were away from the normal to the plates. Therefore, the important parameter for the dolphin may be information level rather than echo level. A third feature of the dolphin sonar system that is often overlooked is the fact that the sonar is mounted on a very flexible and mobile platform. Dolphins conduct sonar searches in an adaptive manner in that the trajectory of the animal at any given time will be the results of echoes received previously. A dolphin will not be restricted to running preprogrammed track lines or transect but is free to maneuver as the situation dictate. Therefore, a dolphin can approach and search on an object at different orientation and obtain whatever information it needs to recognize a target. The dolphin sonar system has evolved over millions of years as nature s way to optimize an important sensory modality. Humans can take advantage of the natural selection process that has been working in dolphins to improve technological sonar. One obvious direction that should be pursued is the use of broadband signals that imitate the signals used by dolphins. There may be a temptation to adopt a longer broadband signal such as FM signals used by some bats in order to project more energy into the water. I caution against such an approach and suggest that more weight should be placed on using natural section as a guide and we should strive to first produce a short-range sonar system that can perform as well as the dolphin. Perhaps, after such a system is developed, tested and used should we can seek to improve on nature. There are many problems that still need to be solved in terms of processing broadband sonar echoes. The manner in which dolphins conduct sonar searches is another area of research that should be pursued. A system in which the sonar echoes dictate the specific trajectory of a mobile platform at any given time needs to be developed. I believe that we can make considerable progress in developing better sonar by following along the path that have been provided by dolphins. 6. References [1] Au, W. W. L.. The Sonar of Dolphins. Springer- Verlag, New York. [2] Norris, K. S. Some problems of echolocation in cetaceans, in Marine Bioacoustics, edited by W. N. Tavolga, Pergamon, N. Y., pp [3] Norris, K. S. Peripheral sound processing in odontocetes, in Animal Sonar System, edited by R. G. Busnel and J. F. Fish, Plenum, N. Y., pp [4] Møhl, B., Au, W. W. L., Pawloski, J., & Nachtigall, P. E. "Dolphin hearing: Relative sensitivity as a function of point of application of a contact sound source in the jaw and head region". J. Acoustic. Soc. Am., Vol. 105: , [5] Johnson, C. S. "Sound detection thresholds in marine mammals". In Marine BioAcoustics W. Tavolga (Ed.). Pergamon, New York, pp , III

8 [6] Au, W. W. L., Lemonds, D. W., Vlachos, S., Nachtigall, P. E., & Roitblat, H. L. "Atlantic bottlenose dolphin hearing threshold for brief broadband signals". Journal of Comparative Pyschology, Vol. 116: , [7] Au, W. W. L., & Moore, P. W. B. "Receiving Beam Patterns and Directivity Indices of the Atlantic Bottlenose Dolphin Tursiops truncatus". J. Acoustic. Soc. Am., Vol. 75, , [8] Lemonds, D. W., Au, W. W. L., Nachtigall, P. E., Roitblat, H. L., and Vlachos, S. A. High frequency auditory filter shapes in an Atlantic bottlenose dolphin, J. Acoust. Soc. Am., Vol. 108, 2000, p.2614(a). [9] Cranford, T. "In Search of Impulse Sound Sources in Odontocetes". In Hearing by Whales and Dolphins W. W. L. Au & A. N. Popper & R. R. Fay (Eds.). Springer-Verlag, New York, pp [10] Nachtigall, P. E. "Odontocete echolocation performance on object size, shape and material". In Animal Sonar Systems R. G. Busnel & J. F. Fish (Eds.). Plenum Press, New York, pp , [11] Murchison, A. E. "Maximum detection range and range resolution in echolocating bottlenose porpoises (Tursiops truncatus)". In Animal Sonar [12] Au, W. W. L., & Snyder, K. J. "Long-range target detection in open waters by an echolocating Atlantic bottlenose dolphin". J. Acoustic. Soc. Am., 56, , [13] Au, W. W. L. "Application of the reverberationlimited form of the sonar equation to dolphin echolocation". J. Acoust. Soc. Am., 92, , 1992 [14] Evans, W. W., & Powell, B. A. "Discrimination of different metallic plates by an echolocating delphinid". In Animal Sonar Systems: Biology and Bionics R. G. Busnel (Ed.). Laboratoire de Physiologie Acoustique, Jouy-en-Josas, pp , [15] Au, W. W. L., and Martin, D. Sonar discrimination of metallic plates by dolphins and humans, in Animal Sonar: Processes and Performance, edited by P. E. Nachtigall and P.W.B. Moore, Plenum Press, N. Y. Pp , [16] Au, W. W. L., & Pawloski, D. A. "Cylinder wall thickness difference discrimination by an echolocating Atlantic bottlenose dolphin". J Comp Physiol A, 172, 41-47, III