Bottlenose dolphin (Tursiops truncatus) detection of simulated echoes from normal and time-reversed clicks James J. Finnerana) U.S. Navy Marine Mammal Program, Space and Naval Warfare Systems Center Pacific, Code 71510, 53560 Hull Street, San Diego, California 92152
Teri Wu, Nancy Borror, and Megan Tormey National Marine Mammal Foundation, 2240 Shelter Island Drive, #200, San Diego, California 92106
Arial Brewer, Amy Black, and Kimberly Bakhtiari G2 Software Systems, 4250 Pacific Highway, Suite 125, San Diego, California 92110
(Received 5 May 2013; revised 13 September 2013; accepted 26 September 2013) In matched filter processing, a stored template of the emitted sonar pulse is compared to echoes to locate individual replicas of the emitted pulse embedded in the echo stream. A number of experiments with bats have suggested that bats utilize matched filter processing for target ranging, but not for target detection. For dolphins, the few available data suggest that dolphins do not utilize matched filter processing. In this study, the effect of time-reversing a dolphin’s emitted click was investigated. If the dolphin relied upon matched filter processing, time-reversal of the click would be expected to reduce the correlation between the (unaltered) click and the echoes and therefore lower detection performance. Two bottlenose dolphins were trained to perform a phantom echo detection task. On a small percentage of trials (“probe trials”), a dolphin’s emitted click was time-reversed before interacting with the phantom echo system. Data from the normal and time-reversed trials were then analyzed and compared. There were no significant differences in detection performance or click emissions between the normal and time-reversed conditions for either subject, suggesting that the dolphins did not utilize matched filter processing for this echo detection task. [http://dx.doi.org/10.1121/1.4824678] PACS number(s): 43.80.Ka, 43.80.Lb [WWA]
I. INTRODUCTION
Bats and echolocating odontocetes possess welldeveloped biological sonar (biosonar) systems that allow them to find objects and navigate through their surroundings. These systems have evolved to solve problems familiar to those facing the designers of hardware sonars: how to detect and classify objects in complicated acoustic environments featuring background noise, reverberation, and clutter. For this reason, there is much interest in understanding how biosonar systems function in order to apply the lessons learned to improve the performance of hardware sonars. Much of what is known about animal biosonar has come from controlled psychophysical experiments conducted with certain species of bats and bottlenose dolphins. The performance of both bats and dolphins during these experiments is often compared to a match-filter or cross correlation processor, which represents a benchmark for evaluating the performance of different types of sonar receivers—an “ideal receiver” (Au, 1993). In matched filter processing, a stored template of the emitted sonar pulse is compared to the incoming echoes to locate individual replicas of the emitted pulse embedded in the echo stream. In a coherent receiver, the cross correlation function between the echo stream and template is used (the phase of echoes is preserved); in a a)
Author to whom correspondence should be addressed. Electronic mail: [email protected]
4548
Pages: 4548–4555
noncoherent receiver, the envelope of the cross correlation function is used (Moss and Schnitzler, 1989; Simmons et al., 1990). A number of psychophysical experiments with bats have suggested that bats utilize matched filter processing for target ranging. The results of early target ranging experiments with bats could be predicted using the envelope of the cross correlation function (Simmons, 1973). Later experiments, utilizing a “jittered-echo” paradigm to reduce the influence of head movement, suggested that big brown bats (Eptesicus fuscus) utilize the fine structure of the cross correlation function, not just the envelope (Simmons, 1979). These data remain controversial: the results were replicated by Simmons et al. (1990) but not by Moss and Schnitzler (1989) or Menne et al. (1986), and alternate explanations for the observed “hyperacuity” for perception of echo delays have been proposed (e.g., Beedholm and Mohl, 1998). Investigators have also studied matched filter processing in bats by examining the effects of altering the relationship between sonar emissions and returning echoes. Alteration of the emitted sonar pulse before it interacted with a target would be expected to reduce the correlation between the signal template (based on the unaltered click) and the returning echoes and therefore degrade the animal’s performance during a biosonar task. Møhl (1986) and Masters and Jacobs (1989) examined the effects of time-reversed echoes on detection thresholds in bats. Both studies showed that detection thresholds were not affected by time-reversing the
J. Acoust. Soc. Am. 134 (6), December 2013 0001-4966/2013/134(6)/4548/8/$30.00
echoes, which contradicts the idea that bats utilize a cross correlation receiver for target detection. Masters and Jacobs (1989) also investigated the effects of echo time reversal on range discrimination. Unlike detection, range discrimination was much worse with time-reversed echoes, suggesting that range discrimination does involve matched filtering. For dolphins, there are few data on the use of matched filter processing. Target detection in noise experiments have indicated that dolphins perform substantially worse than an ideal receiver (Au and Pawloski, 1989) and dolphin range discrimination experiments showed that performance varied with range (Murchison, 1980a). Both of these results imply that dolphins do not utilize matched filter processing; however, specific experiments to investigate jittered echo delay resolution or the effects of time-reversed echoes have not been carried out in dolphins. This paper describes an experiment designed to examine the effects of time-reversing a dolphin’s emitted click on the resulting target detection performance. The methods were similar to those employed by Møhl (1986) and Masters and Jacobs (1989). Two bottlenose dolphins were trained to perform a phantom echo detection task. On a small percentage of trials (“probe trials”), the dolphin’s emitted click was time-reversed before interacting with the phantom echo system. Data from the normal and time-reversed trials were analyzed and compared. If the dolphin relied upon matched filter processing for target detection, time-reversal of the clicks would be expected to reduce the correlation between the (unaltered) click and the echoes and therefore lower detection performance. II. METHODS A. Subjects and test apparatus
Subjects consisted of two bottlenose dolphins: SAY (female, 33 yr, 245 kg), and IND (male, 10 yr, 175 kg). The dolphins were housed in floating netted enclosures measuring 9 m 9 m to 9 m 18 m and located in San Diego Bay, CA. Previous evoked potential hearing tests demonstrated “normal” upper cutoff frequencies (between 120–140 kHz) in both dolphins (e.g., Houser and Finneran, 2006), and both had previously participated in tasks utilizing phantom echoes (e.g., Branstetter et al., 2012; Finneran, 2013). Test sessions were conducted in a 9 m 9 m enclosure modified to contain an observation aperture in the side of the enclosure facing San Diego Bay. The aperture was a 1.8-m square frame constructed of polyvinylchloride pipe and covered with netting except for a 35-cm diameter hoop opening in the center, at a depth of 1 m. The dolphins were trained to position their heads in the hoop opening, so that they faced outward from the enclosure, toward San Diego Bay. The only known reflectors between the aperture and the opposite side of the Bay (approximately 1 km away) consisted of a series of 50-cm square concrete piles, located approximately 30–40 m away. Ambient noise at the test site was dominated by snapping shrimp, other dolphins, and small vessel traffic; average noise pressure spectral density levels were 60 dB re 1 lPa2/Hz at 100 kHz. J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
Two piezoelectric transducers were positioned 1 m from the subject. One of the transducers (1089D, International Transducer Corp, Santa Barbara, CA) acted as a hydrophone, and was used to receive the dolphin’s outgoing echolocation clicks for input to the phantom echo generator (PEG). The second transducer (TC4013, Reson Inc., Slangerup, Denmark) functioned as a sound projector and was used to transmit phantom echoes to the subject.
B. Phantom echo generation
The PEG consisted of a TMS320C6713 digital signal processor (DSP) starter kit (Texas Instruments, Dallas, TX) with an analog input/output daughtercard (AED109, Signalware Corporation, Colorado Springs, CO). The operation of the PEG is shown in Fig. 1. The output from the click receiver hydrophone was high-pass filtered (5 kHz; VP1000, Reson Inc., Slangerup, Denmark), attenuated if necessary (6 dB, custom) and band-pass filtered (5–200 kHz, 8-pole Butterworth; 3C module, Krohn-Hite Corporation, Brockton, MA), then digitized by the AED109 with a 1-MHz sampling rate and 12-bit resolution. The digitized hydrophone signal was then passed to a (software) threshold-crossing click detector. If a click was detected, the click waveform was convolved with a target impulse response, scaled in amplitude, and stored in one of 16 discrete memory buffers. During a small proportion of trials, the click waveform was time-reversed before the convolution was performed. The convolution was performed in the frequency domain and the echo was “pre-equalized” for the transmitting frequency response of the TC4013 projector. Three target impulse responses were used: 0, A, and C. Impulse response 0 represented a target-absent condition—the echo amplitude was set to zero and no echoes were generated. Target A was based on target strength measurements of a die-cast model of the Star Wars “Death Star” (Titanium Series; Hasbro, Pawtucket, RI), a 76.2-mm diameter sphere featuring a 9-mm diameter concave spherical cavity on one side (Finneran et al., 2010). Target C featured a unit impulse response, so the echoes matched the received click. [This target was designated as target C, rather than target B, to avoid confusion with the target B used in previous studies (Finneran et al., 2010; Finneran, 2013)]. Target A was included as a more complex impulse response than target C. After time delays appropriate for the simulated target range, the contents of each echo memory buffer were converted to analog at 1 MHz with 12-bit resolution using the AED109. The analog echo signal was then filtered (5–200 kHz, 8-pole Butterworth; 3C module, Krohn-Hite Corporation, Brockton, MA), attenuated if necessary (PA4, Tucker-Davis Technologies, Alachua, FL), amplified (Pro 5200, Crest Audio, Meridian, MS), and used to drive the echo transmitter. The PEG also generated a second analog signal, termed the “echo sync pulse,” that was time-locked with the echo waveform and had a peak amplitude proportional to the click p-p amplitude. The dolphin clicks and echo sync pulses were digitized at 2 MHz and 16-bit resolution by an NI PXIe 6368 multifunction data acquisition Finneran et al.: Time-reversal of dolphin biosonar clicks
4549
FIG. 1. (Color online) Operating principles of the phantom echo generator (PEG).
device (National Instruments, Austin, TX) and stored on hard disk for later analysis.
C. Test sequence
The dolphins were trained to echolocate while in the hoop and report the presence of phantom echoes by producing an acoustic response (a “whistle”). Sixty-percent of the trials were echo-present trials, where the PEG began producing echoes in response to dolphin clicks 0–3 s after the experimenter initiated the trial. The remaining trials were echo-absent, where no echoes were produced and the dolphin was required to remain in the hoop and withhold the whistle response for up to 7 s. The period of time from 0 to 4 s after the onset of the echoes was designated as the response interval. If the dolphin whistled during a response interval (a hit) or withheld the response for an entire echoabsent trial (a correct rejection), it was rewarded with fish (one capelin). A whistle response outside of a response interval (a false alarm) or a failure to respond during a response interval (a miss) resulted in the dolphin being recalled to the surface with no fish reward. If a dolphin did not echolocate during a trial, stopped echolocating before the appearance of echoes, or was visually observed to be echolocating on another object, the dolphin was recalled and the trial was ignored and not further analyzed. Data were first collected over a period of several weeks using echo C, then data were collected with echo A. Each test session consisted of 120 trials. Sixty of the trials featured a 25-m simulated target range; these trials provided the data of interest. The remaining 60 trials were included to maintain subject motivation and attentiveness and were evenly divided among ranges of 50, 100, and 150 m. The range was randomized from one trial to the next. Of the 25-m trials, six from each session (four echo-present, two echo-absent, 10% of the 25-m trials, 5% of all trials) were probe trials, where 4550
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
the dolphin click was time-reversed before convolution with the target impulse response. Dolphins were always recalled after probe trials with no reward (i.e., responses to probe trials were never reinforced). The relative echo level, RL, was defined as the difference between the received p-p sound pressure level (SPL) of the echo (LE) and the p-p SPL of the dolphin’s click measured at 1 m (LC): RL ¼ LE – LC. Four RLs were used: 82, 87, 92, and 97 dB; these specific values were determined after preliminary testing to estimate the subjects’ echo detection thresholds. The RL was randomized across the session for each target range, with an equal number of trials at each RL. During each session, there was typically one probe trial at each RL. At least 20 probe trials were conducted for each combination of target and RL. D. Data analysis
Custom software was used to analyze the clicks from the hydrophone recording and to determine the time latency between the onset of echo reception and the time of the subject’s whistle response. The p-p SPL, sound exposure level (SEL), and center frequency of each click were also calculated and averaged across all echo-present trials with the same RL value. The probability of detection, P(D), was defined as the number of hits divided by the number of echopresent trials, and the false alarm rate, P(FA), was the number of false alarms divided by the number of echo-absent trials. Nonlinear regression (Graphpad Software, 2003) was used to fit the equation PðDÞ ¼
1 1 þ 10
S ðRL50 RLÞ
;
(1)
where S and RL50 are fitting parameters, to the P(D) vs RL data for the normal and time-reversed click data. Equation (1) is a form of a standard four-parameter logistic equation Finneran et al.: Time-reversal of dolphin biosonar clicks
with a variable slope, with the bottom and top parameters constrained to 0 and 1, respectively. The parameters RL50 and S define the relative echo level and slope, respectively, when P(D) ¼ 0.5. Curve-fits utilized the individual trial data, with misses coded with a value of 0 and hits coded with a value of 1. After curve-fitting, an F-test was performed to determine if the fitting parameters differed between the normal and time-reversed data for each subject; i.e., were the curves relating P(D) and RL different between the normal and time-reversed click data (Graphpad Software, 2003)? III. RESULTS
Table I provides the number of data collection sessions conducted for each dolphin and the resulting numbers of echo-present trials and clicks analyzed for each combination of target condition and RL. Figure 2 shows representative examples of clicks produced by IND and SAY and the resulting echoes based on normal and time-reversed clicks. Clicks from both dolphins resembled exponentially damped sinusoids with center frequencies near 80–90 kHz. Figure 2 also shows the envelope of the response of a matched filter to the echoes from normal and time-reversed trials (the envelope of the cross correlation between click and echo, Au, 1993). Figure 3 shows the mean click SPLs (upper panels) and center frequencies (lower panels) utilized by IND and SAY for each target type (A or C) and condition (normal or timereversed). Data are shown for only those trials resulting in correct detections. Click SPLs tended to be higher for echo A in both dolphins, though the differences were generally small (2–4 dB). Center frequencies were also higher in both dolphins for echo A. There was little change in click level or center frequency with RL or between the normal and time-reversed conditions. Across all target conditions for both subjects, mean click SELs varied from 142 to 153 dB re 1 lPa2s (dB SEL hereafter) and the difference between the TABLE I. Numbers of sessions, trials, and analyzed clicks for each subject and target condition. S þ trialsnumber of echo-present trials. The two numbers indicate the values for the normal/time-reversed data. Subject IND
SAY
Target
Sessions
RL (dB)
Sþ trials
Clicks
A
22
C
25
A
23
C
23
82 87 92 97 82 87 92 97 82 87 92 97 82 87 92 97
162/20 157/20 162/20 171/20 170/22 166/21 170/21 164/21 169/20 166/20 168/20 159/20 185/23 183/23 184/23 181/24
703/107 769/81 769/119 815/117 628/143 772/88 1152/112 1029/145 647/89 708/68 820/124 775/124 978/112 1252/188 1097/152 719/187
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
FIG. 2. (Color online) Examples of clicks, echoes, and envelopes of matched-filter outputs for each signal. Normal trial—test condition with normal clicks; time-reversed trial—test condition with time-reversed clicks.
numeric values for mean p-p SPL (in dB re 1 lPa) and mean SEL (in dB SEL) was nearly constant: 57.3–58.6, very close to the value of 58 cited by Au (1993) for dolphins in Kaneohe Bay. Figure 4 shows the (median) response latencies (upper panels) for IND and SAY for echo A and C and the normal and time-reversed trials. The response latency for each trial was defined as the time between reception of the first echo and the onset of the subject’s response. Since the inter-click interval and two-way travel time will affect the response latency, the total number of echoes received before the subject’s response is also shown in Fig. 4 (lower panels). Both the response latency and number of echoes increased with decreasing RL and followed exponential patterns. Response latencies and number of echoes were highly variable, especially for the time-reversed data, which were based on few samples, and for the lower RLs. For both subjects, median response latencies at the highest RL values were near 400–500 ms and the required number of echoes was approximately five. For each combination of RL and target type, the differences between the medians for the normal and timereversed trials were small relative to the dispersion of the data. Figure 5 shows the P(D) and P(FA) for IND and SAY for each condition. False alarm rates were generally low, from 1% to 6% for both subjects. All P(D) data were fit reasonably well using Eq. (1): the (adjusted) goodness of fit parameters comparing the curve-fits to the calculated P(D) at each RL were between 0.809 and 0.999. There were no significant differences between the normal and time-reversed datasets for either IND or SAY (Table II), and the differences between the detection thresholds based on a P(D) ¼ 0.75 were less than 2 dB for all test conditions (Table II). Finneran et al.: Time-reversal of dolphin biosonar clicks
4551
FIG. 3. (Color online) Click p-p SPL (upper panels) and click center frequencies (lower panels) for IND and SAY for each target condition (A) and (C) and trial type (normal and timereversed). Symbols represent the mean values and the error bars show the standard deviation.
FIG. 4. (Color online) Response latency (upper panels) and the number of echoes required (lower panels) for IND and SAY for each target condition (A) and (C) and trial type (normal and time-reversed). Symbols represent the medians and the error bars show the inter-quartile distances. For visualization, the response latencies were fit with the Pieron function: t0 þ b I a, where I is the ratio of echo mean-square pressure to click mean-square pressure (I ¼ 10RL/10), and a, b, and t0 are fitting parameters. The number of echoes data were fit with exponential decay functions.
4552
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
Finneran et al.: Time-reversal of dolphin biosonar clicks
FIG. 5. (Color online) Probability of detection, P(D), for IND and SAY with targets A and C. The lines are the best-fits of Eq. (1) to the data. The shaded regions represent the 95% confidence intervals for the normal (light shading) and time-reversed conditions (darker shading). The false alarm rates, P(FA), for each subject and target are listed in each panel.
IV. DISCUSSION A. Dolphin click emissions and performance
Both dolphins utilized stereotypical clicks resembling exponentially decaying sinusoids, similar to on-axis recordings of clicks from other dolphins (Au, 1993). The ratio of click SPL to SEL was also very close to that reported previously (Au, 1993), suggesting that, in the absence of direct measurements, SELs (in dB SEL) for on-axis bottlenose dolphin clicks may be estimated from p-p SPLs (in dB re 1 lPa) by subtracting 58, regardless of the specific environment. Although both dolphins utilized higher mean source levels for echo A, there was little difference in click level with relative echo level. Previous echolocation studies utilizing phantom echoes with variable range (i.e., echo delay) and RL reported increasing click SPLs with decreasing RL (e.g., Supin et al., 2011; Supin et al., 2012), therefore the failure of either animal to increase click levels as RL decreased and
P(D) decreased is somewhat surprising. It may be that the dolphins were already operating near their maximum source levels and that no further increase could be accomplished (see Au and Penner, 1981). Detection thresholds (in terms of RL) for each subject were similar for the two targets: 93 to 95 dB for IND and 93 to 90 dB for SAY. IND’s thresholds were slightly lower than SAY, but the differences were relatively small (2–5 dB). Using the RL at threshold and the mean click SELs, the echo SELs at threshold may be estimated as 51–56 dB SEL for IND and 60–61 dB SEL for SAY. Using an estimated ambient noise spectral density of 60 dB re 1 lPa2/Hz and receiving directivity index of 20 dB, the ratios of echo sound exposure at threshold to noise spectral density (analogous to the Ee/N0 parameter defined by Au, 1993) were found to be approximately 13 dB for IND and 20 dB for SAY. These values are higher than those previously estimated for dolphins performing detection tasks in Kaneohe
TABLE II. Detection thresholds for IND and SAY for each target condition (A) or (C) and trial type (normal, time-reversed). Thresholds are based on the relative echo level (RL) for a 75% probability of detection. Detection threshold (RL, dB) Subject IND IND SAY SAY
Target
Normal
Time-reversed
Difference
Fit parameters different?
A C A C
93.1 95.2 90.8 89.0
93.0 95.1 92.6 90.4
0.1 0.1 1.8 1.4
No, F(2, 728) ¼ 0.00854, p ¼ 0.992 No, F(2, 751) ¼ 2.093, p ¼ 0.124 No, F(2, 738) ¼ 0.5380, p ¼ 0.584 No, F(2, 728) ¼ 2.870, p ¼ 0.0572
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
Finneran et al.: Time-reversal of dolphin biosonar clicks
4553
Bay: 7 dB (Murchison, 1980b), 10 dB (Au and Snyder, 1980), 4–9 dB (Au and Penner, 1981), and 7 dB (Turl et al., 1987). The higher detection thresholds in the present study could be a result of the dolphins’ relatively conservative response bias (i.e., low false alarm rates). The concrete piles located 30 to 40 m in front of the dolphins may have also affected performance for the 25-m simulated range condition. Finally, previous work has shown differences in click emissions and auditory evoked potentials between ordered and random target presentations (Supin et al., 2012). Previous echolocation studies presenting detection thresholds utilized a fixed range during a session (Murchison, 1980b; Au and Penner, 1981; Turl et al., 1987) or an ordered increase in range between blocks of trials (Au and Snyder, 1980). It is possible that the randomized target range and RL conditions resulted in lower detection performance for the subjects, compared to what they could have achieved with a fixed range or ordered (i.e., predictable) changes in target range and/or RL. Previous dolphin biosonar studies defined response latency or reaction time as the time between the last click emission and the onset of the response (Au et al., 1982; Au and Turl, 1983; Ridgway et al., 2012). This definition is somewhat problematic, since dolphins may respond before the cessation of clicking, leading to negative reaction times (Ridgway et al., 2012). The definition used in the present study (first echo onset to response onset) more closely matches the traditional definition of simple reaction time: the time interval from stimulus onset to response onset (Luce, 1986). This is similar to the “total latency” metric used by Au et al. (1982). Reaction times are not normally distributed, but rather have a sharp rise at small latency values and a longer, decaying “tail” at the larger latency values (Whelan, 2008). For this reason, median latencies with the interquartile distance were reported in the present study. The median latencies increased as RL decreased, meaning latency increased as task difficulty increased, as expected. Reaction times for IND tended to be smaller than for SAY at the same condition; this may reflect the higher detectability of the echoes for IND (thresholds for IND were slightly lower), individual differences in the time required to produce their whistle response, or be related to age or gender (Noble et al., 1964). Differences between median reactions times were similar between the normal and time-reversed conditions; however, the sample sizes for the normal and time-reversed trials were not the same, so comparisons across conditions are biased [the time-reversed data, with fewer samples, will tend to have larger medians (Whelan, 2008), as evidenced in Fig. 4]. B. Effect of time-reversal
There were no significant differences in detection performance between the normal and time-reversed conditions for either subject. The dolphins’ click emissions were also similar for the normal and time-reversed conditions. Given the nature of the task and the probe trial paradigm, it seems unlikely that the subjects would have been able to dynamically adjust an internal click “template” for each individual 4554
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
trial. Therefore, the data suggest that the dolphin subjects did not utilize matched filter processing for this echo detection task. This finding supports earlier work in dolphins, where echo detection in noise and range resolution experiments showed dolphin performance to not match predictions for a matched filter processor. Results of the present study are identical to earlier findings for bats, where detection performance was not affected by time-reversing the echoes (Møhl, 1986; Masters and Jacobs, 1989). Given the short duration and impulsive, broadband nature of the dolphin clicks, it is perhaps not surprising that no effect was observed from the time-reversal. Comparison of the normal and time-reversed signals in Fig. 2 shows mainly subtle differences between the two conditions, especially for echo C. For echo A, there were differences in the latencies of the components of the matched filter output that appear to be well within the 0.5–0.6 ls time difference discrimination thresholds reported for dolphins (Au and Pawloski, 1992). However, the primary components of the matched filter outputs are very similar across the normal and time-reversed conditions. It may be possible that a dolphin could discriminate between the two echoes at high signalto-noise ratios, even though detection thresholds are not significantly different. Although time-reversal of echoes did not affect detection performance in bats, it did have a dramatic effect on echo-ranging performance (Masters and Jacobs, 1989). Comparable experiments have not been conducted with dolphins and would be an important step to further comparisons of biosonar system operation between bats and dolphins. ACKNOWLEDGMENTS
The authors wish to thank Randall Dear and Jim Powell for logistic support and Dorian Houser, Patrick Moore, Jason Mulsow, and Brian Branstetter for helpful discussions on the experimental approach and data. The study followed a protocol approved by the Institutional Animal Care and Use Committee at the Space and Naval Warfare Systems Center (SSC) Pacific and the Navy Bureau of Medicine and Surgery, and followed all applicable U.S. Department of Defense guidelines. Financial support was provided by the SSC Pacific Naval Innovative Science and Engineering (NISE) program.
Au, W. L., and Snyder, K. J. (1980). “Long-range target detection in open waters by an echolocating Atlantic bottlenosed dolphin (Tursiops truncatus),” J. Acoust. Soc. Am. 68, 1077–1084. Au, W. W. L. (1993). The Sonar of Dolphins (Springer–Verlag, New York), 227 pp. Au, W. W. L., and Pawloski, D. A. (1989). “A comparison of signal detection between an echolocating dolphin and an optimal receiver,” J. Comp. Physiol. A 164, 451–458. Au, W. W. L., and Pawloski, D. A. (1992). “Cylinder wall thickness difference discrimination by an echolocating Atlantic bottlenose dolphin,” J. Comp. Physiol. A 170, 41–47. Au, W. W. L., and Penner, R. H. (1981). “Target detection in noise by echolocating Atlantic bottlenose dolphins,” J. Acoust. Soc. Am. 70, 687–693. Au, W. W. L., Penner, R. H., and Kadane, J. (1982). “Acoustic behavior of echolocating Atlantic Bottlenose Dolphins,” J. Acoust. Soc. Am. 71, 1269–1275. Finneran et al.: Time-reversal of dolphin biosonar clicks
Au, W. W. L., and Turl, C. W. (1983). “Target detection in reverberation by an echolocating Atlantic bottlenosed dolphin (Tursiops truncatus),” J. Acoust. Soc. Am. 73, 1676–1681. Beedholm, K., and Mohl, B. (1998). “Bat sonar: an alternative interpretation of the 10-ns jitter result,” J. Comp. Physiol. A 182, 259–266. Branstetter, B. K., Finneran, J. J., Fletcher, E. A., Weisman, B. C., and Ridgway, S. H. (2012). “Dolphins can maintain vigilant behavior through echolocation for 15 days without interruption or cognitive impairment,” PLoS ONE 7, 1–10. Finneran, J. J. (2013). “Dolphin ‘packet’ use during long-range echolocation tasks,” J. Acoust. Soc. Am. 133, 1796–1810. Finneran, J. J., Houser, D. S., Moore, P. W., Branstetter, B. K., Trickey, J. S., and Ridgway, S. H. (2010). “A method to enable a bottlenose dolphin (Tursiops truncatus) to echolocate while out of water,” J. Acoust. Soc. Am. 128, 1483–1489. GraphPad Software (2003). “GraphPad Prism,” (GraphPad Software, San Diego, CA). Houser, D. S., and Finneran, J. J. (2006). “Variation in the hearing sensitivity of a dolphin population obtained through the use of evoked potential audiometry,” J. Acoust. Soc. Am. 120, 4090–4099. Luce, R. D. (1986). Response Times: Their Role in Inferring Elementary Mental Organization (Oxford University Press, New York), 577 pp. Masters, W. M., and Jacobs, S. C. (1989). “Target detection and range resolution by the big brown bat (Eptesicus fuscus),” J. Comp. Physiol. A 166, 65–73. Menne, D., and Hackbarth, H. (1986). “Accuracy of distance measurement in the bat Eptesicus fuscus: theoretical aspects and computer simulations,” J. Acoust. Soc. Am. 79, 386–397. Møhl, B. (1986). “Detection by a pipistrelle bat of normal and reversed replica of its sonar pulses,” Acustica 61, 75–82. Moss, C. F., and Schnitzler, H.-U. (1989). “Accuracy of target ranging in echolocating bats: acoustic information processing,” J. Comp. Physiol. A 165, 383–393.
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
Murchison, A. E. (1980a). “Detection range and range resolution of echolocating Bottlenose Porpoise (Tursiops truncatus),” in Animal Sonar Systems, edited by R. G. Busnel and J. F. Fish (Plenum, New York), pp. 43–70. Murchison, A. E. (1980b). “Maximum detection range and range resolution in echolocating bottlenose porpoises, Tursiops truncatus (Montagu),” Ph.D. Dissertation, University of California Santa Cruz, 284 pp. Noble, C., Baker, B. L., and Jones, T. A. (1964). “Age and sex parameters in psychomotor learning,” Percept. Mot. Skills 19, 935–945. Ridgway, S. H., Elsberry, W. R., Blackwood, D. J., Kamolnick, T., Todd, M., Carder, D. A., Chaplin, M., and Cranford, T. W. (2012). “Vocal reporting of echolocation targets: Dolphins often report before click trains end,” J. Acoust. Soc. Am. 131, 593–598. Simmons, J. A. (1973). “The resolution of target range by echolocating bats,” J. Acoust. Soc. Am. 54, 157–173. Simmons, J. A. (1979). “Perception of echo phase information in bat sonar,” Science 204, 1336–1338. Simmons, J. A., Ferragamo, M., Moss, C. F., Stevenson, S. B., and Altes, R. A. (1990). “Discrimination of jittered sonar echoes by the echolocating bat, Eptesicus fuscus: The shape of target images in echolocation,” J. Comp. Physiol. A 167, 589–616. Supin, A. Y., Nachtigall, P. E., and Breese, M. (2011). “Interaction of emitted sonar pulses and simulated echoes in a false killer whale: An evokedpotential study,” J. Acoust. Soc. Am. 130, 1711–1720. Supin, A. Y., Nachtigall, P. E., and Breese, M. (2012). “A whale better adjusts the biosonar to ordered rather than to random changes in the echo parameters,” J. Acoust. Soc. Am. 132, 1811–1819. Turl, C. W., Penner, R. H., and Au, W. W. L. (1987). “Comparison of target detection capabilities of the beluga and bottlenose dolphin,” J. Acoust. Soc. Am. 82, 1487–1491. Whelan, R. (2008). “Effective analysis of reaction time data,” Psychol. Rec. 58, 475–482.
Finneran et al.: Time-reversal of dolphin biosonar clicks
4555
Copyright of Journal of the Acoustical Society of America is the property of American Institute of Physics and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Teri Wu, Nancy Borror, and Megan Tormey National Marine Mammal Foundation, 2240 Shelter Island Drive, #200, San Diego, California 92106
Arial Brewer, Amy Black, and Kimberly Bakhtiari G2 Software Systems, 4250 Pacific Highway, Suite 125, San Diego, California 92110
(Received 5 May 2013; revised 13 September 2013; accepted 26 September 2013) In matched filter processing, a stored template of the emitted sonar pulse is compared to echoes to locate individual replicas of the emitted pulse embedded in the echo stream. A number of experiments with bats have suggested that bats utilize matched filter processing for target ranging, but not for target detection. For dolphins, the few available data suggest that dolphins do not utilize matched filter processing. In this study, the effect of time-reversing a dolphin’s emitted click was investigated. If the dolphin relied upon matched filter processing, time-reversal of the click would be expected to reduce the correlation between the (unaltered) click and the echoes and therefore lower detection performance. Two bottlenose dolphins were trained to perform a phantom echo detection task. On a small percentage of trials (“probe trials”), a dolphin’s emitted click was time-reversed before interacting with the phantom echo system. Data from the normal and time-reversed trials were then analyzed and compared. There were no significant differences in detection performance or click emissions between the normal and time-reversed conditions for either subject, suggesting that the dolphins did not utilize matched filter processing for this echo detection task. [http://dx.doi.org/10.1121/1.4824678] PACS number(s): 43.80.Ka, 43.80.Lb [WWA]
I. INTRODUCTION
Bats and echolocating odontocetes possess welldeveloped biological sonar (biosonar) systems that allow them to find objects and navigate through their surroundings. These systems have evolved to solve problems familiar to those facing the designers of hardware sonars: how to detect and classify objects in complicated acoustic environments featuring background noise, reverberation, and clutter. For this reason, there is much interest in understanding how biosonar systems function in order to apply the lessons learned to improve the performance of hardware sonars. Much of what is known about animal biosonar has come from controlled psychophysical experiments conducted with certain species of bats and bottlenose dolphins. The performance of both bats and dolphins during these experiments is often compared to a match-filter or cross correlation processor, which represents a benchmark for evaluating the performance of different types of sonar receivers—an “ideal receiver” (Au, 1993). In matched filter processing, a stored template of the emitted sonar pulse is compared to the incoming echoes to locate individual replicas of the emitted pulse embedded in the echo stream. In a coherent receiver, the cross correlation function between the echo stream and template is used (the phase of echoes is preserved); in a a)
Author to whom correspondence should be addressed. Electronic mail: [email protected]
4548
Pages: 4548–4555
noncoherent receiver, the envelope of the cross correlation function is used (Moss and Schnitzler, 1989; Simmons et al., 1990). A number of psychophysical experiments with bats have suggested that bats utilize matched filter processing for target ranging. The results of early target ranging experiments with bats could be predicted using the envelope of the cross correlation function (Simmons, 1973). Later experiments, utilizing a “jittered-echo” paradigm to reduce the influence of head movement, suggested that big brown bats (Eptesicus fuscus) utilize the fine structure of the cross correlation function, not just the envelope (Simmons, 1979). These data remain controversial: the results were replicated by Simmons et al. (1990) but not by Moss and Schnitzler (1989) or Menne et al. (1986), and alternate explanations for the observed “hyperacuity” for perception of echo delays have been proposed (e.g., Beedholm and Mohl, 1998). Investigators have also studied matched filter processing in bats by examining the effects of altering the relationship between sonar emissions and returning echoes. Alteration of the emitted sonar pulse before it interacted with a target would be expected to reduce the correlation between the signal template (based on the unaltered click) and the returning echoes and therefore degrade the animal’s performance during a biosonar task. Møhl (1986) and Masters and Jacobs (1989) examined the effects of time-reversed echoes on detection thresholds in bats. Both studies showed that detection thresholds were not affected by time-reversing the
J. Acoust. Soc. Am. 134 (6), December 2013 0001-4966/2013/134(6)/4548/8/$30.00
echoes, which contradicts the idea that bats utilize a cross correlation receiver for target detection. Masters and Jacobs (1989) also investigated the effects of echo time reversal on range discrimination. Unlike detection, range discrimination was much worse with time-reversed echoes, suggesting that range discrimination does involve matched filtering. For dolphins, there are few data on the use of matched filter processing. Target detection in noise experiments have indicated that dolphins perform substantially worse than an ideal receiver (Au and Pawloski, 1989) and dolphin range discrimination experiments showed that performance varied with range (Murchison, 1980a). Both of these results imply that dolphins do not utilize matched filter processing; however, specific experiments to investigate jittered echo delay resolution or the effects of time-reversed echoes have not been carried out in dolphins. This paper describes an experiment designed to examine the effects of time-reversing a dolphin’s emitted click on the resulting target detection performance. The methods were similar to those employed by Møhl (1986) and Masters and Jacobs (1989). Two bottlenose dolphins were trained to perform a phantom echo detection task. On a small percentage of trials (“probe trials”), the dolphin’s emitted click was time-reversed before interacting with the phantom echo system. Data from the normal and time-reversed trials were analyzed and compared. If the dolphin relied upon matched filter processing for target detection, time-reversal of the clicks would be expected to reduce the correlation between the (unaltered) click and the echoes and therefore lower detection performance. II. METHODS A. Subjects and test apparatus
Subjects consisted of two bottlenose dolphins: SAY (female, 33 yr, 245 kg), and IND (male, 10 yr, 175 kg). The dolphins were housed in floating netted enclosures measuring 9 m 9 m to 9 m 18 m and located in San Diego Bay, CA. Previous evoked potential hearing tests demonstrated “normal” upper cutoff frequencies (between 120–140 kHz) in both dolphins (e.g., Houser and Finneran, 2006), and both had previously participated in tasks utilizing phantom echoes (e.g., Branstetter et al., 2012; Finneran, 2013). Test sessions were conducted in a 9 m 9 m enclosure modified to contain an observation aperture in the side of the enclosure facing San Diego Bay. The aperture was a 1.8-m square frame constructed of polyvinylchloride pipe and covered with netting except for a 35-cm diameter hoop opening in the center, at a depth of 1 m. The dolphins were trained to position their heads in the hoop opening, so that they faced outward from the enclosure, toward San Diego Bay. The only known reflectors between the aperture and the opposite side of the Bay (approximately 1 km away) consisted of a series of 50-cm square concrete piles, located approximately 30–40 m away. Ambient noise at the test site was dominated by snapping shrimp, other dolphins, and small vessel traffic; average noise pressure spectral density levels were 60 dB re 1 lPa2/Hz at 100 kHz. J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
Two piezoelectric transducers were positioned 1 m from the subject. One of the transducers (1089D, International Transducer Corp, Santa Barbara, CA) acted as a hydrophone, and was used to receive the dolphin’s outgoing echolocation clicks for input to the phantom echo generator (PEG). The second transducer (TC4013, Reson Inc., Slangerup, Denmark) functioned as a sound projector and was used to transmit phantom echoes to the subject.
B. Phantom echo generation
The PEG consisted of a TMS320C6713 digital signal processor (DSP) starter kit (Texas Instruments, Dallas, TX) with an analog input/output daughtercard (AED109, Signalware Corporation, Colorado Springs, CO). The operation of the PEG is shown in Fig. 1. The output from the click receiver hydrophone was high-pass filtered (5 kHz; VP1000, Reson Inc., Slangerup, Denmark), attenuated if necessary (6 dB, custom) and band-pass filtered (5–200 kHz, 8-pole Butterworth; 3C module, Krohn-Hite Corporation, Brockton, MA), then digitized by the AED109 with a 1-MHz sampling rate and 12-bit resolution. The digitized hydrophone signal was then passed to a (software) threshold-crossing click detector. If a click was detected, the click waveform was convolved with a target impulse response, scaled in amplitude, and stored in one of 16 discrete memory buffers. During a small proportion of trials, the click waveform was time-reversed before the convolution was performed. The convolution was performed in the frequency domain and the echo was “pre-equalized” for the transmitting frequency response of the TC4013 projector. Three target impulse responses were used: 0, A, and C. Impulse response 0 represented a target-absent condition—the echo amplitude was set to zero and no echoes were generated. Target A was based on target strength measurements of a die-cast model of the Star Wars “Death Star” (Titanium Series; Hasbro, Pawtucket, RI), a 76.2-mm diameter sphere featuring a 9-mm diameter concave spherical cavity on one side (Finneran et al., 2010). Target C featured a unit impulse response, so the echoes matched the received click. [This target was designated as target C, rather than target B, to avoid confusion with the target B used in previous studies (Finneran et al., 2010; Finneran, 2013)]. Target A was included as a more complex impulse response than target C. After time delays appropriate for the simulated target range, the contents of each echo memory buffer were converted to analog at 1 MHz with 12-bit resolution using the AED109. The analog echo signal was then filtered (5–200 kHz, 8-pole Butterworth; 3C module, Krohn-Hite Corporation, Brockton, MA), attenuated if necessary (PA4, Tucker-Davis Technologies, Alachua, FL), amplified (Pro 5200, Crest Audio, Meridian, MS), and used to drive the echo transmitter. The PEG also generated a second analog signal, termed the “echo sync pulse,” that was time-locked with the echo waveform and had a peak amplitude proportional to the click p-p amplitude. The dolphin clicks and echo sync pulses were digitized at 2 MHz and 16-bit resolution by an NI PXIe 6368 multifunction data acquisition Finneran et al.: Time-reversal of dolphin biosonar clicks
4549
FIG. 1. (Color online) Operating principles of the phantom echo generator (PEG).
device (National Instruments, Austin, TX) and stored on hard disk for later analysis.
C. Test sequence
The dolphins were trained to echolocate while in the hoop and report the presence of phantom echoes by producing an acoustic response (a “whistle”). Sixty-percent of the trials were echo-present trials, where the PEG began producing echoes in response to dolphin clicks 0–3 s after the experimenter initiated the trial. The remaining trials were echo-absent, where no echoes were produced and the dolphin was required to remain in the hoop and withhold the whistle response for up to 7 s. The period of time from 0 to 4 s after the onset of the echoes was designated as the response interval. If the dolphin whistled during a response interval (a hit) or withheld the response for an entire echoabsent trial (a correct rejection), it was rewarded with fish (one capelin). A whistle response outside of a response interval (a false alarm) or a failure to respond during a response interval (a miss) resulted in the dolphin being recalled to the surface with no fish reward. If a dolphin did not echolocate during a trial, stopped echolocating before the appearance of echoes, or was visually observed to be echolocating on another object, the dolphin was recalled and the trial was ignored and not further analyzed. Data were first collected over a period of several weeks using echo C, then data were collected with echo A. Each test session consisted of 120 trials. Sixty of the trials featured a 25-m simulated target range; these trials provided the data of interest. The remaining 60 trials were included to maintain subject motivation and attentiveness and were evenly divided among ranges of 50, 100, and 150 m. The range was randomized from one trial to the next. Of the 25-m trials, six from each session (four echo-present, two echo-absent, 10% of the 25-m trials, 5% of all trials) were probe trials, where 4550
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
the dolphin click was time-reversed before convolution with the target impulse response. Dolphins were always recalled after probe trials with no reward (i.e., responses to probe trials were never reinforced). The relative echo level, RL, was defined as the difference between the received p-p sound pressure level (SPL) of the echo (LE) and the p-p SPL of the dolphin’s click measured at 1 m (LC): RL ¼ LE – LC. Four RLs were used: 82, 87, 92, and 97 dB; these specific values were determined after preliminary testing to estimate the subjects’ echo detection thresholds. The RL was randomized across the session for each target range, with an equal number of trials at each RL. During each session, there was typically one probe trial at each RL. At least 20 probe trials were conducted for each combination of target and RL. D. Data analysis
Custom software was used to analyze the clicks from the hydrophone recording and to determine the time latency between the onset of echo reception and the time of the subject’s whistle response. The p-p SPL, sound exposure level (SEL), and center frequency of each click were also calculated and averaged across all echo-present trials with the same RL value. The probability of detection, P(D), was defined as the number of hits divided by the number of echopresent trials, and the false alarm rate, P(FA), was the number of false alarms divided by the number of echo-absent trials. Nonlinear regression (Graphpad Software, 2003) was used to fit the equation PðDÞ ¼
1 1 þ 10
S ðRL50 RLÞ
;
(1)
where S and RL50 are fitting parameters, to the P(D) vs RL data for the normal and time-reversed click data. Equation (1) is a form of a standard four-parameter logistic equation Finneran et al.: Time-reversal of dolphin biosonar clicks
with a variable slope, with the bottom and top parameters constrained to 0 and 1, respectively. The parameters RL50 and S define the relative echo level and slope, respectively, when P(D) ¼ 0.5. Curve-fits utilized the individual trial data, with misses coded with a value of 0 and hits coded with a value of 1. After curve-fitting, an F-test was performed to determine if the fitting parameters differed between the normal and time-reversed data for each subject; i.e., were the curves relating P(D) and RL different between the normal and time-reversed click data (Graphpad Software, 2003)? III. RESULTS
Table I provides the number of data collection sessions conducted for each dolphin and the resulting numbers of echo-present trials and clicks analyzed for each combination of target condition and RL. Figure 2 shows representative examples of clicks produced by IND and SAY and the resulting echoes based on normal and time-reversed clicks. Clicks from both dolphins resembled exponentially damped sinusoids with center frequencies near 80–90 kHz. Figure 2 also shows the envelope of the response of a matched filter to the echoes from normal and time-reversed trials (the envelope of the cross correlation between click and echo, Au, 1993). Figure 3 shows the mean click SPLs (upper panels) and center frequencies (lower panels) utilized by IND and SAY for each target type (A or C) and condition (normal or timereversed). Data are shown for only those trials resulting in correct detections. Click SPLs tended to be higher for echo A in both dolphins, though the differences were generally small (2–4 dB). Center frequencies were also higher in both dolphins for echo A. There was little change in click level or center frequency with RL or between the normal and time-reversed conditions. Across all target conditions for both subjects, mean click SELs varied from 142 to 153 dB re 1 lPa2s (dB SEL hereafter) and the difference between the TABLE I. Numbers of sessions, trials, and analyzed clicks for each subject and target condition. S þ trialsnumber of echo-present trials. The two numbers indicate the values for the normal/time-reversed data. Subject IND
SAY
Target
Sessions
RL (dB)
Sþ trials
Clicks
A
22
C
25
A
23
C
23
82 87 92 97 82 87 92 97 82 87 92 97 82 87 92 97
162/20 157/20 162/20 171/20 170/22 166/21 170/21 164/21 169/20 166/20 168/20 159/20 185/23 183/23 184/23 181/24
703/107 769/81 769/119 815/117 628/143 772/88 1152/112 1029/145 647/89 708/68 820/124 775/124 978/112 1252/188 1097/152 719/187
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
FIG. 2. (Color online) Examples of clicks, echoes, and envelopes of matched-filter outputs for each signal. Normal trial—test condition with normal clicks; time-reversed trial—test condition with time-reversed clicks.
numeric values for mean p-p SPL (in dB re 1 lPa) and mean SEL (in dB SEL) was nearly constant: 57.3–58.6, very close to the value of 58 cited by Au (1993) for dolphins in Kaneohe Bay. Figure 4 shows the (median) response latencies (upper panels) for IND and SAY for echo A and C and the normal and time-reversed trials. The response latency for each trial was defined as the time between reception of the first echo and the onset of the subject’s response. Since the inter-click interval and two-way travel time will affect the response latency, the total number of echoes received before the subject’s response is also shown in Fig. 4 (lower panels). Both the response latency and number of echoes increased with decreasing RL and followed exponential patterns. Response latencies and number of echoes were highly variable, especially for the time-reversed data, which were based on few samples, and for the lower RLs. For both subjects, median response latencies at the highest RL values were near 400–500 ms and the required number of echoes was approximately five. For each combination of RL and target type, the differences between the medians for the normal and timereversed trials were small relative to the dispersion of the data. Figure 5 shows the P(D) and P(FA) for IND and SAY for each condition. False alarm rates were generally low, from 1% to 6% for both subjects. All P(D) data were fit reasonably well using Eq. (1): the (adjusted) goodness of fit parameters comparing the curve-fits to the calculated P(D) at each RL were between 0.809 and 0.999. There were no significant differences between the normal and time-reversed datasets for either IND or SAY (Table II), and the differences between the detection thresholds based on a P(D) ¼ 0.75 were less than 2 dB for all test conditions (Table II). Finneran et al.: Time-reversal of dolphin biosonar clicks
4551
FIG. 3. (Color online) Click p-p SPL (upper panels) and click center frequencies (lower panels) for IND and SAY for each target condition (A) and (C) and trial type (normal and timereversed). Symbols represent the mean values and the error bars show the standard deviation.
FIG. 4. (Color online) Response latency (upper panels) and the number of echoes required (lower panels) for IND and SAY for each target condition (A) and (C) and trial type (normal and time-reversed). Symbols represent the medians and the error bars show the inter-quartile distances. For visualization, the response latencies were fit with the Pieron function: t0 þ b I a, where I is the ratio of echo mean-square pressure to click mean-square pressure (I ¼ 10RL/10), and a, b, and t0 are fitting parameters. The number of echoes data were fit with exponential decay functions.
4552
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
Finneran et al.: Time-reversal of dolphin biosonar clicks
FIG. 5. (Color online) Probability of detection, P(D), for IND and SAY with targets A and C. The lines are the best-fits of Eq. (1) to the data. The shaded regions represent the 95% confidence intervals for the normal (light shading) and time-reversed conditions (darker shading). The false alarm rates, P(FA), for each subject and target are listed in each panel.
IV. DISCUSSION A. Dolphin click emissions and performance
Both dolphins utilized stereotypical clicks resembling exponentially decaying sinusoids, similar to on-axis recordings of clicks from other dolphins (Au, 1993). The ratio of click SPL to SEL was also very close to that reported previously (Au, 1993), suggesting that, in the absence of direct measurements, SELs (in dB SEL) for on-axis bottlenose dolphin clicks may be estimated from p-p SPLs (in dB re 1 lPa) by subtracting 58, regardless of the specific environment. Although both dolphins utilized higher mean source levels for echo A, there was little difference in click level with relative echo level. Previous echolocation studies utilizing phantom echoes with variable range (i.e., echo delay) and RL reported increasing click SPLs with decreasing RL (e.g., Supin et al., 2011; Supin et al., 2012), therefore the failure of either animal to increase click levels as RL decreased and
P(D) decreased is somewhat surprising. It may be that the dolphins were already operating near their maximum source levels and that no further increase could be accomplished (see Au and Penner, 1981). Detection thresholds (in terms of RL) for each subject were similar for the two targets: 93 to 95 dB for IND and 93 to 90 dB for SAY. IND’s thresholds were slightly lower than SAY, but the differences were relatively small (2–5 dB). Using the RL at threshold and the mean click SELs, the echo SELs at threshold may be estimated as 51–56 dB SEL for IND and 60–61 dB SEL for SAY. Using an estimated ambient noise spectral density of 60 dB re 1 lPa2/Hz and receiving directivity index of 20 dB, the ratios of echo sound exposure at threshold to noise spectral density (analogous to the Ee/N0 parameter defined by Au, 1993) were found to be approximately 13 dB for IND and 20 dB for SAY. These values are higher than those previously estimated for dolphins performing detection tasks in Kaneohe
TABLE II. Detection thresholds for IND and SAY for each target condition (A) or (C) and trial type (normal, time-reversed). Thresholds are based on the relative echo level (RL) for a 75% probability of detection. Detection threshold (RL, dB) Subject IND IND SAY SAY
Target
Normal
Time-reversed
Difference
Fit parameters different?
A C A C
93.1 95.2 90.8 89.0
93.0 95.1 92.6 90.4
0.1 0.1 1.8 1.4
No, F(2, 728) ¼ 0.00854, p ¼ 0.992 No, F(2, 751) ¼ 2.093, p ¼ 0.124 No, F(2, 738) ¼ 0.5380, p ¼ 0.584 No, F(2, 728) ¼ 2.870, p ¼ 0.0572
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
Finneran et al.: Time-reversal of dolphin biosonar clicks
4553
Bay: 7 dB (Murchison, 1980b), 10 dB (Au and Snyder, 1980), 4–9 dB (Au and Penner, 1981), and 7 dB (Turl et al., 1987). The higher detection thresholds in the present study could be a result of the dolphins’ relatively conservative response bias (i.e., low false alarm rates). The concrete piles located 30 to 40 m in front of the dolphins may have also affected performance for the 25-m simulated range condition. Finally, previous work has shown differences in click emissions and auditory evoked potentials between ordered and random target presentations (Supin et al., 2012). Previous echolocation studies presenting detection thresholds utilized a fixed range during a session (Murchison, 1980b; Au and Penner, 1981; Turl et al., 1987) or an ordered increase in range between blocks of trials (Au and Snyder, 1980). It is possible that the randomized target range and RL conditions resulted in lower detection performance for the subjects, compared to what they could have achieved with a fixed range or ordered (i.e., predictable) changes in target range and/or RL. Previous dolphin biosonar studies defined response latency or reaction time as the time between the last click emission and the onset of the response (Au et al., 1982; Au and Turl, 1983; Ridgway et al., 2012). This definition is somewhat problematic, since dolphins may respond before the cessation of clicking, leading to negative reaction times (Ridgway et al., 2012). The definition used in the present study (first echo onset to response onset) more closely matches the traditional definition of simple reaction time: the time interval from stimulus onset to response onset (Luce, 1986). This is similar to the “total latency” metric used by Au et al. (1982). Reaction times are not normally distributed, but rather have a sharp rise at small latency values and a longer, decaying “tail” at the larger latency values (Whelan, 2008). For this reason, median latencies with the interquartile distance were reported in the present study. The median latencies increased as RL decreased, meaning latency increased as task difficulty increased, as expected. Reaction times for IND tended to be smaller than for SAY at the same condition; this may reflect the higher detectability of the echoes for IND (thresholds for IND were slightly lower), individual differences in the time required to produce their whistle response, or be related to age or gender (Noble et al., 1964). Differences between median reactions times were similar between the normal and time-reversed conditions; however, the sample sizes for the normal and time-reversed trials were not the same, so comparisons across conditions are biased [the time-reversed data, with fewer samples, will tend to have larger medians (Whelan, 2008), as evidenced in Fig. 4]. B. Effect of time-reversal
There were no significant differences in detection performance between the normal and time-reversed conditions for either subject. The dolphins’ click emissions were also similar for the normal and time-reversed conditions. Given the nature of the task and the probe trial paradigm, it seems unlikely that the subjects would have been able to dynamically adjust an internal click “template” for each individual 4554
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
trial. Therefore, the data suggest that the dolphin subjects did not utilize matched filter processing for this echo detection task. This finding supports earlier work in dolphins, where echo detection in noise and range resolution experiments showed dolphin performance to not match predictions for a matched filter processor. Results of the present study are identical to earlier findings for bats, where detection performance was not affected by time-reversing the echoes (Møhl, 1986; Masters and Jacobs, 1989). Given the short duration and impulsive, broadband nature of the dolphin clicks, it is perhaps not surprising that no effect was observed from the time-reversal. Comparison of the normal and time-reversed signals in Fig. 2 shows mainly subtle differences between the two conditions, especially for echo C. For echo A, there were differences in the latencies of the components of the matched filter output that appear to be well within the 0.5–0.6 ls time difference discrimination thresholds reported for dolphins (Au and Pawloski, 1992). However, the primary components of the matched filter outputs are very similar across the normal and time-reversed conditions. It may be possible that a dolphin could discriminate between the two echoes at high signalto-noise ratios, even though detection thresholds are not significantly different. Although time-reversal of echoes did not affect detection performance in bats, it did have a dramatic effect on echo-ranging performance (Masters and Jacobs, 1989). Comparable experiments have not been conducted with dolphins and would be an important step to further comparisons of biosonar system operation between bats and dolphins. ACKNOWLEDGMENTS
The authors wish to thank Randall Dear and Jim Powell for logistic support and Dorian Houser, Patrick Moore, Jason Mulsow, and Brian Branstetter for helpful discussions on the experimental approach and data. The study followed a protocol approved by the Institutional Animal Care and Use Committee at the Space and Naval Warfare Systems Center (SSC) Pacific and the Navy Bureau of Medicine and Surgery, and followed all applicable U.S. Department of Defense guidelines. Financial support was provided by the SSC Pacific Naval Innovative Science and Engineering (NISE) program.
Au, W. L., and Snyder, K. J. (1980). “Long-range target detection in open waters by an echolocating Atlantic bottlenosed dolphin (Tursiops truncatus),” J. Acoust. Soc. Am. 68, 1077–1084. Au, W. W. L. (1993). The Sonar of Dolphins (Springer–Verlag, New York), 227 pp. Au, W. W. L., and Pawloski, D. A. (1989). “A comparison of signal detection between an echolocating dolphin and an optimal receiver,” J. Comp. Physiol. A 164, 451–458. Au, W. W. L., and Pawloski, D. A. (1992). “Cylinder wall thickness difference discrimination by an echolocating Atlantic bottlenose dolphin,” J. Comp. Physiol. A 170, 41–47. Au, W. W. L., and Penner, R. H. (1981). “Target detection in noise by echolocating Atlantic bottlenose dolphins,” J. Acoust. Soc. Am. 70, 687–693. Au, W. W. L., Penner, R. H., and Kadane, J. (1982). “Acoustic behavior of echolocating Atlantic Bottlenose Dolphins,” J. Acoust. Soc. Am. 71, 1269–1275. Finneran et al.: Time-reversal of dolphin biosonar clicks
Au, W. W. L., and Turl, C. W. (1983). “Target detection in reverberation by an echolocating Atlantic bottlenosed dolphin (Tursiops truncatus),” J. Acoust. Soc. Am. 73, 1676–1681. Beedholm, K., and Mohl, B. (1998). “Bat sonar: an alternative interpretation of the 10-ns jitter result,” J. Comp. Physiol. A 182, 259–266. Branstetter, B. K., Finneran, J. J., Fletcher, E. A., Weisman, B. C., and Ridgway, S. H. (2012). “Dolphins can maintain vigilant behavior through echolocation for 15 days without interruption or cognitive impairment,” PLoS ONE 7, 1–10. Finneran, J. J. (2013). “Dolphin ‘packet’ use during long-range echolocation tasks,” J. Acoust. Soc. Am. 133, 1796–1810. Finneran, J. J., Houser, D. S., Moore, P. W., Branstetter, B. K., Trickey, J. S., and Ridgway, S. H. (2010). “A method to enable a bottlenose dolphin (Tursiops truncatus) to echolocate while out of water,” J. Acoust. Soc. Am. 128, 1483–1489. GraphPad Software (2003). “GraphPad Prism,” (GraphPad Software, San Diego, CA). Houser, D. S., and Finneran, J. J. (2006). “Variation in the hearing sensitivity of a dolphin population obtained through the use of evoked potential audiometry,” J. Acoust. Soc. Am. 120, 4090–4099. Luce, R. D. (1986). Response Times: Their Role in Inferring Elementary Mental Organization (Oxford University Press, New York), 577 pp. Masters, W. M., and Jacobs, S. C. (1989). “Target detection and range resolution by the big brown bat (Eptesicus fuscus),” J. Comp. Physiol. A 166, 65–73. Menne, D., and Hackbarth, H. (1986). “Accuracy of distance measurement in the bat Eptesicus fuscus: theoretical aspects and computer simulations,” J. Acoust. Soc. Am. 79, 386–397. Møhl, B. (1986). “Detection by a pipistrelle bat of normal and reversed replica of its sonar pulses,” Acustica 61, 75–82. Moss, C. F., and Schnitzler, H.-U. (1989). “Accuracy of target ranging in echolocating bats: acoustic information processing,” J. Comp. Physiol. A 165, 383–393.
J. Acoust. Soc. Am., Vol. 134, No. 6, December 2013
Murchison, A. E. (1980a). “Detection range and range resolution of echolocating Bottlenose Porpoise (Tursiops truncatus),” in Animal Sonar Systems, edited by R. G. Busnel and J. F. Fish (Plenum, New York), pp. 43–70. Murchison, A. E. (1980b). “Maximum detection range and range resolution in echolocating bottlenose porpoises, Tursiops truncatus (Montagu),” Ph.D. Dissertation, University of California Santa Cruz, 284 pp. Noble, C., Baker, B. L., and Jones, T. A. (1964). “Age and sex parameters in psychomotor learning,” Percept. Mot. Skills 19, 935–945. Ridgway, S. H., Elsberry, W. R., Blackwood, D. J., Kamolnick, T., Todd, M., Carder, D. A., Chaplin, M., and Cranford, T. W. (2012). “Vocal reporting of echolocation targets: Dolphins often report before click trains end,” J. Acoust. Soc. Am. 131, 593–598. Simmons, J. A. (1973). “The resolution of target range by echolocating bats,” J. Acoust. Soc. Am. 54, 157–173. Simmons, J. A. (1979). “Perception of echo phase information in bat sonar,” Science 204, 1336–1338. Simmons, J. A., Ferragamo, M., Moss, C. F., Stevenson, S. B., and Altes, R. A. (1990). “Discrimination of jittered sonar echoes by the echolocating bat, Eptesicus fuscus: The shape of target images in echolocation,” J. Comp. Physiol. A 167, 589–616. Supin, A. Y., Nachtigall, P. E., and Breese, M. (2011). “Interaction of emitted sonar pulses and simulated echoes in a false killer whale: An evokedpotential study,” J. Acoust. Soc. Am. 130, 1711–1720. Supin, A. Y., Nachtigall, P. E., and Breese, M. (2012). “A whale better adjusts the biosonar to ordered rather than to random changes in the echo parameters,” J. Acoust. Soc. Am. 132, 1811–1819. Turl, C. W., Penner, R. H., and Au, W. W. L. (1987). “Comparison of target detection capabilities of the beluga and bottlenose dolphin,” J. Acoust. Soc. Am. 82, 1487–1491. Whelan, R. (2008). “Effective analysis of reaction time data,” Psychol. Rec. 58, 475–482.
Finneran et al.: Time-reversal of dolphin biosonar clicks
4555
Copyright of Journal of the Acoustical Society of America is the property of American Institute of Physics and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
