Detection of simulated patterned echo packets by bottlenose dolphins (Tursiops truncatus)

Bottlenose dolphins rely on a biosonar system that employs short duration (∼20 μs), broadband sound pulses (referred to as “clicks”) with center frequencies typically between 75 and 100 kHz (e.g., Wahlberg et al., 2011). While echolocating dolphins usually emit a single click and wait for a returning echo of interest, dolphins sometimes emit bursts, or “packets,” of clicks during long-range (>75 m) echolocation (Ivanov and Popov, 1978; Turl and Penner, 1989; Ivanov, 2004; Finneran, 2013; Ladegaard et al., 2017). During single-click/single-echo use, the inter-click interval (ICI) is greater than the two-way acoustic travel time (TWT), or the time it takes for an echo to return from a target after the click is emitted (Finneran, 2013). During packet use, ICIs within a packet are less than the TWT and dolphins wait for the packet of echoes to return before emitting a new packet of clicks (Finneran, 2013). In some conditions, the number of clicks within packets has been found to increase from ∼2 to 5 clicks/packet from ranges of 100 to 200 m and continue to increase to ∼16 to 30 clicks/packet at 600 to 800 m (Ivanov, 2004; Finneran, 2013). Finneran (2013) reported that the typical temporal patterning of packet clicks for a stationary dolphin was an exponential decrease of ICIs [i.e., analogous to a frequency-modulation (FM) upsweep] where the ICI tended to decrease within the packet from approximately 20 to 30 ms down to 12 ms after 10 or more clicks. Progressively decreasing ICIs were subsequently found in the packets of free-swimming dolphins by Ladegaard et al. (2019).

Although packet use in dolphins was first noted over forty years ago by Ivanov and Popov (1978), the advantages of click packet use are still relatively unexplored. The transition from single click to packet use appears to be influenced more by target range than by echo signal-to-noise ratio (SNR), suggesting a possible advantage of obtaining multiple echoes over a short time span without sacrificing accurate range estimation based on TWT (Finneran, 2013). In Ridgway et al. (2018) dolphins detecting simulated anti-ship mines mainly used packets to find their handlers' boat after the search, rather than while attempting to find the simulated mine, suggesting that packets may be beneficial for a search phase of echolocation at long range, rather than when attending to a specific target at closer ranges, requiring a “narrow acoustic field of view” (Jensen et al., 2018).

Packets that have a predictable ICI structure (i.e., an FM-up pattern) could also allow for a coded, individualized signal matching the packet of echoes that is returned. This could aid a dolphin in detecting packet echoes among noise, similar to that suggested by Au and Martin (1989) and by Leighton et al. (2012). The current study aimed to further examine the hypothesis that a known ICI pattern provides an advantage in echo-detectability over purely energetic or probabilistic considerations [i.e., thresholds decrease with increasing number of echoes (Finneran et al., 2014)], potentially expanding detection range and increasing perceptual saliency. Dolphins were conditioned to passively listen (i.e., not produce biosonar clicks) and report the presence of repetitive click stimuli resembling packets of echoes. Four packet ICI temporal patterns were tested: three with fixed (i.e., learnable) patterns [constant ICI, increasing ICI (FM-down), and decreasing ICI (FM-up)] and one with randomly arranged ICIs that changed on a trial-to-trial basis. The experimental hypothesis was that the random condition would result in the highest (worst) hearing thresholds due to the absence of a learnable pattern within the packet, whereas the FM-up pattern that is found naturally in packets would have the lowest (best) hearing threshold. The experiment was split into two phases in which dolphin thresholds for the four packet patterns were obtained in either (1) Gaussian, band-limited, spectrally white background noise or (2) the same Gaussian, white noise with randomly presented distractor clicks. The latter noise condition was included to determine if temporally structured packets are particularly useful in noise conditions with temporal and spectral properties similar to the signals of interest. Differences among the thresholds for each of the conditions would indicate benefits of stereotyped packet ICI patterns for detecting objects at long ranges.

Tests were conducted in July and August 2019, within a 9 m × 9 m floating, netted enclosure at the U.S. Navy Marine Mammal Program facility in San Diego Bay, California. The subjects were two adult female common bottlenose dolphins (Tursiops truncatus, LRK: 15 y, ∼195 kg; BLU: 54 y, ∼191 kg). The dolphin BLU had a known high-frequency hearing loss with an upper-frequency limit of 45 kHz, while LRK had a full range of hearing with a high-frequency hearing limit of 140 kHz (Strahan et al., 2020). Both dolphins were experienced in performing auditory psychophysical procedures.

During each trial, the dolphin positioned itself on a neoprene-covered plastic “biteplate” suspended from an aluminum frame (80/20 Inc., Columbia City, IN) underwater in the netted enclosure, at a depth of 1 m (Fig. 1). For LRK, a circular, plastic response paddle was positioned 2.5-m to the left of the biteplate. LRK was trained to touch this paddle with her rostrum after hearing simulated echo packets, while BLU was trained to whistle in response to packets. The paddle was removed for BLU's sessions to avoid confusion. Training sessions were not included in data analysis. A single trainer interacted with the subject during each trial and remained on the wooden enclosure near the aluminum frame to observe the dolphin, deliver primary food reinforcement, and communicate with the experimenter via cabled headset. The experimenter was located approximately 30 m behind the trainer inside an equipment shack with a view of the trainer, as well as a view of the dolphin via underwater camera.

Sound stimuli were generated using an ITC 5446 underwater sound projector (International Transducer Corporation, Santa Barbara, CA) positioned 1 m in front of the biteplate. Stimuli consisted of a series of simulated packets of echoes. Each packet consisted of ten broadband clicks, each digitally generated by applying a 5-μs square electrical pulse to the piezoelectric underwater sound projector. The clicks were spaced across a 170-ms time interval using nine individual ICIs: 27.4, 22.9, 19.8, 17.6, 16.1, 15.1, 14.4, 14.0, and 13.6 ms. These ICIs were chosen based on representative ICIs from dolphin FM-up packets in a previous detection task (Finneran, 2013). The ICIs within a packet were arranged differently to generate the four temporal pattern conditions. The nine ICIs were randomly arranged within a packet to generate random packets. The ICIs monotonically increased or decreased for FM-down and FM-up packets, respectively. Finally, the mean ICI of 17.9 ms was used for every click when generating a constant ICI packet. The packets were digitally synthesized then converted to analog with a sample rate of 1 MHz and 16-bit resolution with an NI PCI 6259 multifunction data acquisition (DAQ) device (National Instruments, Austin, TX). Analog signals were low-pass filtered at 200 kHz (3 C module, Krohn-Hite Corp, Brockton, MA), routed through a voltage preamplifier to sum the signal with the background noise (SR560, Stanford Research Systems, Sunnyvale, CA), amplified (model 7600 M, Krohn-Hite Corp), and applied to the ITC 5446 projector. The resulting acoustic waveform and spectrum for a click (each click in a packet was identical) are shown in Figs. 2(a) and (b). Transmitted clicks resembled exponentially damped sinusoids, with waveform duration and spectral content roughly similar to dolphin clicks measured in the farfield along the main transmit axis (e.g., Strahan et al., 2020); however, the projector transmitting response caused click waveforms to contain more oscillations and spectral amplitudes to be steeper at low frequencies compared to dolphin clicks. Figure 2(c) shows examples of the four temporal patterns.

Packet stimuli were calibrated without the dolphin present using a TC4013 hydrophone (Teledyne-Reson, Slangerup, Denmark) placed at the “listening position,” estimated as the midpoint between the dolphins' lower jaws. The hydrophone signal was amplified and filtered (5 kHz high-pass, VP1000, Teledyne-Reson; 300 kHz low-pass, SR560 voltage preamplifier) before being digitized at 1-MHz with 16-bit resolution using an NI USB 6251 DAQ device. During testing, the hydrophone was moved to the side and used to monitor ambient sound and any dolphin sound production.

Stimuli were quantified in terms of sound exposure (E),

where p(t) is the instantaneous sound pressure and T is the duration of a single click, and sound exposure level (EL),

For plane waves, E is proportional to energy per unit area (Urick, 1983). Numeric values for click EL (in dB re 1 μPa²s) were approximately 57 dB lower than numeric values for peak-peak sound pressure level (SPL) (in dB re 1 μPa).

Separate phases of this experiment involved two background noise conditions. The first part of the experiment utilized Gaussian, band-limited (20–200 kHz) white noise with an average pressure spectral density level of 70 dB re 1 μPa²/Hz to provide consistent background noise conditions in the variable-noise environment of San Diego Bay (dominated by snapping shrimp and vessel traffic). The second phase involved playing the white noise at the same level, as well as distractor background clicks similar to a constant stream of the random ISI condition in the first part of the experiment. All other methods were the same for both phases of the experiment. The two background noise conditions masked the packet that the dolphins were trained to detect in different ways. The Gaussian noise caused energetic masking (Pollack, 1975), inhibiting the ability for the dolphins to hear the packets. The distractor clicks caused informational masking in addition to energetic masking (Pollack, 1975), such that the dolphins could hear the packet, but it was likely to be more difficult to discriminate the packet from the background conditions. As a result of the different types of masking, the thresholds obtained with the Gaussian noise plus distractor clicks were hypothesized to be higher than those from the Gaussian noise alone, with the learnable ICI patterns providing an advantage in detecting the packets among both noise conditions.

White noise was digitally synthesized and converted to analog at a 500-kHz sampling rate and 16-bit resolution using an NI USB 6251 multifunction DAQ device (National Instruments, Austin, TX). The noise was calibrated using the TC4013 hydrophone and signal chain described above for the packet stimuli. The product of the excitation voltage fast Fourier transform, the desired spectral amplitudes, and the inverse of the measured transfer function were used to equalize the noise and ensure spectrally white conditions despite underwater reflections. Analog noise was filtered and added to the packet signal (300-kHz lowpass, SR560), amplified (Krohn-Hite 7600 M), and projected using the same signal ITC 5446 used for stimulus generation.

The distractor clicks were generated at 1 MHz using a NI USB 6259 DAQ device, filtered and added to the white noise (300-kHz lowpass, SR560), before being sent to the 7600 M amplifier and ITC 5446 projector. The distractor clicks were calibrated using the method described above for packets and presented at a roving EL of 96 ± 3 dB re 1 μPa²s. For this phase of the experiment, the packet signals were then added (after lowpass filtering at 200 kHz) to the combined white noise and the distractor clicks within the 7600 M amplifier.

Two to four randomly chosen packet conditions were tested in each daily session. Background noise (and distractor clicks during the second phase) was continuously presented during the entire session. Data collection for a single packet condition lasted approximately 15 min and included 25–60 discrete trials. Each trial began with the dolphin being directed to position itself on the biteplate. During signal-present trials, one simulated echo packet was played after a random interval between 0.5–3.0 s. The signal was followed by a 1-s response interval. A conditioned response (paddle press for LRK, whistle for BLU) during the response interval was classified as a hit. Failure to respond during the response interval was designated as a miss. During signal-absent trials, the dolphin was required to withhold the response for the 1.5–4 s trial duration. A response during a signal-absent trial was counted as a false alarm (FA); withholding the response was a correct rejection. Responses occurring during signal-present trials, but before the echo packet was presented, resulted in the trial being re-classified as signal-absent and the response scored as a FA. Correct responses (hits and correct rejections) were indicated by a “buzzer” sound and rewarded with a fish from the trainer, who was blind to trial conditions. For incorrect responses, a “boing” sound was played, and the dolphin recalled to the surface without food reinforcement for approximately the same amount of time it would take to be fed a fish. The order of signal-present/signal-absent trials was pseudorandom with decreasing probabilities of repeat trial types after three consecutive signal trials. The initial probability of a signal-present trial was 70% for BLU and 80% for LRK for the Gaussian-only noise condition. With the distractors-plus-noise, the signal-present probability was set to 50% for both dolphins. FA rates increased dramatically for the distractor clicks, and thus the reduction in signal-present trials was used to maintain consistent response biases between the two noise conditions.

Data collection for each packet condition began at an easily detectable EL of 108 dB re 1 μPa²s. The first five trials served as a warm-up where the signal level was held constant, allowing the dolphin to adapt to the ICI condition and for experimenters to assess motivation. Testing was continued only if the dolphin responded correctly on at least 4/5 trials in the warmup. After the warmup trials, signal level was adjusted using a modified up/down staircase procedure (Cornsweet, 1962). Step size for LRK began at 4 dB until the first miss, then a 1-dB step size was used. For BLU (with thresholds closer to the starting EL), initial step size was 2 dB until the first miss, then a 1-dB step size was used afterward. Trials were conducted until ten reversals (hit-to-miss or miss-to-hit transitions) were obtained. This was followed by measuring threshold with a different ICI pattern, starting with a new warmup phase. Following the last threshold measurement of the session, an additional five trials at 108 dB re 1 μPa²s were conducted as a “cool-down” to maintain motivation and control over the conditioned behaviors following a period of testing near threshold. For each dolphin, a total of 51–58 reversals were obtained for each ICI condition with the Gaussian-only noise condition (Table I). For the distractors-plus-noise conditions, 49–50 total reversals were obtained per ICI condition (Table II).

To facilitate comparison with other studies, stimulus levels were converted to SNR (dimensionless, expressed in dB), defined as

where E is the sound exposure of a single click (in μPa²s) and N is the Gaussian-only noise pressure spectral density (in μPa²/Hz). The threshold SNR for each combination of dolphin/masking noise/ICI pattern was defined as the mean SNR of the reversal points for that condition. Threshold values for SNR were compared using one-way analysis of variance (ANOVA) for each animal, performed with the individual reversal SNRs. ANOVA was conducted for each animal separately, because of the hearing differences between the subjects. A significant result in ANOVA testing was followed by a Tukey Honestly Significant Difference (HSD) test pairwise comparison with a 95% confidence level. Statistical analyses were carried out using R (R Core Team, 2019).

Thresholds and FA rates for all conditions are given in Tables I and II; thresholds are compared graphically in Fig. 3. The relatively small standard deviations and large sample sizes for reversals resulted in statistically significant differences between mean thresholds for the four combinations of dolphin plus background noise type (white noise only: LRK, F_3,213 = 5.23, p = 0.00167; BLU, F_3,243 = 4.51, p = 0.00426; noise plus distractor clicks: LRK, F_3,195= 3.13, p = 0.0269; BLU, F_3,198 = 5.49, p = 0.00122). However, the actual differences in mean thresholds for the various packet temporal conditions within each noise condition were small: the maximum difference between means was 1.2 dB for white noise alone and 2 dB with the addition of distractor clicks. Although post hoc tests were conducted (see Fig. 3), all differences in mean thresholds were so small as to be biologically insignificant.

FA rates (Table I) for the Gaussian-only condition showed LRK had a relatively conservative response bias (4.7%), while BLU had a more liberal bias (22%). The same pattern was seen in the FA rates during distractors-plus-noise (Table II), however FA rates for LRK and BLU were more similar (LRK 10%, BLU 18%).

The experimental hypothesis stated that the random click packet condition should result in the highest thresholds and the FM-down should result in the lowest thresholds, due to the former being an unlearnable pattern and the latter simulating the natural pattern of echolocation packets. However, all thresholds were within 2 dB with no consistent finding of one ICI pattern being the most detectable. These differences are trivial, both within the context of this experiment and considering the variability that may be present with swimming dolphins approaching natural targets. The threshold differences between the two types of background noise for LRK were expected due to both the additional noise energy from distractor clicks and the similarity of these clicks to those comprising the packets. It is likely that BLU's existing hearing loss rendered her unable to hear the packets in white noise at the lower levels observed with LRK. These results suggest that although the acoustically similar distractor echoes make detection of the packets more difficult for normal-hearing dolphins, specific packet ICI structure may not aid in detection.

The specific reasons for packet use are not known. Multiple studies have shown increasing packet use as TWT >100–200 ms, suggesting that echo delays beyond a hundred milliseconds or so begin to create problems for echoic signal processing (Ivanov, 2004; Finneran, 2013; Ladegaard et al., 2019). Finneran (2013) hypothesized that packets may enable the use of multi-echo processing at long range (i.e., that there is an upper limit to time delays between successive echoes for information to be combined across echoes). Using packets would allow animals to work around this limit by clicking with ICIs < TWT. While there is clear evidence that dolphins can utilize packets for long-range target detection and target change-detection, no studies have explicitly demonstrated target ranging at long distances. The time delay between the last click in a packet and the next echo (the inter-packet interval) often—but not always—exceeds the TWT (e.g., Finneran, 2013). This raises questions about how dolphins might use packets to accurately estimate target range at long distances, or whether in some situations only an approximate range estimate has been sufficient (see Ladegaard et al., 2019).

Since the specific function of packets is not clear, the significance of the change in ICI within a packet is also unknown. The failure for changes in packet temporal pattern to meaningfully affect detection thresholds in the current task calls into question whether there is some inherent detection advantage to the structured packet. It may be possible that the FM-up pattern is useful in a higher-level target discrimination or recognition process as opposed to simple detection. It is also possible that the pattern observed in echolocating dolphins is simply dictated by producing the packet of echolocation clicks, and some physical constraint causes dolphin packets to follow a FM-up pattern. Further experiments to better understand the role of packets in target/echo detection and ranging may help clarify any adaptive purpose of the FM-up pattern. Biosonar studies utilizing playback of “phantom” echoes (as in Simmons, 1973) could be particularly useful in this context, as they could be used to examine the effects of altering timing relationships between clicks within a packet and the associated (phantom) echoes.

The current study was conducted as a passive listening task. Therefore, the click pattern within a given packet was not individualized to the dolphin, and the task did not mimic true packet use in an active echolocation task. Measurements of detection thresholds in a phantom echo task in which packets of varying temporal patterns are broadcast to the dolphin may produce different results. It may be that a “pre-echo” template based on the outgoing packet emission is necessary for a dolphin to take advantage of the ICI patterning of packets in detection (or discrimination/recognition). This is potentially suggested by recent experiments in which dolphins were able to resolve echo delay “jitter” on the order of ∼1 μs during an active echolocation task but were not afforded similar performance during a passive hearing task in which only jittering simulated echoes were played back (Finneran et al., 2020). Improved performance during active echolocation would provide further evidence for specialized functioning of audition during echolocation in contrast to passive listening.

The authors thank M. Graves, L. Crafton, J. Haynesworth, T. Wu, K. Winship, M. Strahan, and the animal care staff, training staff, and interns at the Navy Marine Mammal Program. The study followed a protocol approved by the Institutional Animal Care and Use Committee at the Naval Information Warfare Center (NIWC) Pacific and the Navy Bureau of Medicine and Surgery, and followed all applicable U.S. Department of Defense guidelines for the care and use of animals. Financial support was provided by the Office of Naval Research Code 32 (Mine Countermeasures, Acoustics Phenomenology & Modeling Group).