When a bat echolocates towards a real target, the delay and amplitude of the echo change together, dependent on the distance of the target. Here we show that FM bats are very sensitive to modulations in amplitude of echoes: Phyllostomus discolor bats were well able to distinguish a virtual target with constant target strength from a virtual target whose target strength was modulated over time. To the best of our knowledge, this study presents the first evidence of an echolocating FM bat detecting a target movement based solely on the modulation of echo-amplitude.
The capability of FM bats to perceive amplitude modulations in echo sequences does not come as a complete surprise (Griffin 1958; Roeder 1963). But our results are the first to experimentally support the implications of theoretical studies on perception of motion with FM echolocation. In those studies, fluttering insects were ensonified with a sequence of synthetic FM signals, and information on wing beat cycle could be decoded from the resulting sequence of echoes (Kober and Schnitzler 1990; Moss and Zagaeski 1994).
The neural adaptations to process echo-amplitude information have been established for the FM bat Myotis lucifugus (Condon et al. 1994): neurons in the inferior colliculus faithfully represent amplitude modulations imposed on a sequence of artificial echolocation calls up to modulation rates around 100 Hz. However, there are three fundamental differences between the electrophysiological experiment by Condon et al. (1994) and the current psychophysical experiment. First, we varied modulation depth to determine a perceptual threshold while Condon et al. always applied 100% (0 dB) modulation depth. Second, the carrier was always a periodic pulse train, whereas in our experiments, the bats themselves determined the ensonification pattern, producing pulses in strobe groups rather than periodically. Third, the combination of pulse rate and modulation rate was always chosen to deliberately avoid a wagon wheel effect in the physiological studies. In summary, this electrophysiological study investigated the effect of amplitude modulation on the neural representation of echo sequences, but the limitations of the electrophysiological protocol preclude a direct comparison with the current psychophysical data.
P. discolor bats were much more sensitive to higher modulation rates than to lower modulation rates: for the lowest modulation rate of 5 Hz, the bats needed around − 10 dB modulation depth to discriminate the constant target from the modulated one, whereas they still detected modulations around − 20 dB modulation depth at the highest modulation rate of 500 Hz (Fig. 3). Strikingly, our findings that sensitivity increased with increasing modulation rate are inconsistent with two related findings: First, they differ from what we found about sensitivity to modulation of echo delay (Baier and Wiegrebe this issue). This suggests that time-variant information on delay and on amplitude of echoes is processed differently. Second, the results are the opposite of what has been found in studies where subjects passively listened to amplitude-modulated sounds (Fay and Wilber 1989). With broadband carrier stimuli, modulation transfer functions in these experiments are always low-pass, i.e., subjects are less sensitive to high modulation rates than to low modulation rates. This is in direct contrast to the results of the current, active echolocation study. We will discuss both points in detail after addressing the bats’ acoustic signals.
As expected from the delay study (Baier and Wiegrebe this issue), individual bats maintained the same call parameters throughout the experiment (Fig. 4). In fact, call parameters were identical to the ones used by the bats in the delay study: we found call durations around 0.4 ms and inter-call intervals (ICIs) around 29 ms. We propose that call parameters were tied to the here simulated target distance of 72 cm, although the bats were not in a formal target-approach situation, where a systematic decrease in ICI and call duration occurs (Griffin et al. 1960). We can ask whether changing the bat’s perceived target distance will result in different call parameters and lead to a different performance in the detection task. Echo sound pressure level (SPL) influences the performance in a range detection task (Denzinger and Schnitzler 1994, 1998). The sensitivity of neurons to sinusoidal amplitude modulation (SAM) is linked to the sound level of the carrier tone (Schuller 1979; Ostwald et al. 1988). Since bats could move freely in our experimental setup, we did not attempt to measure echo SPL. However, we had designed our virtual targets as plain reflectors that very likely yielded target strengths large enough to not affect the modulation detection. The question whether our chosen target range did so remains open.
Comparison to echo-delay modulation
In the companion paper, we have investigated sensitivity for the modulation of echo delay (Baier and Wiegrebe this issue). We found that P. discolor bats were very sensitive to low and high delay-modulation rates and much less so to intermediate delay-modulation rates around 20 and 50 Hz. We proposed that for delay-modulation rates of 20 Hz and higher, bats suffer from an auditory wagon-wheel effect because their call repetition rate matches the modulation rate of the target or an integer multiple thereof. In these cases, the difference in echo delay between consecutive echoes becomes undetectably small. The use of spectral cues that only occur at high modulation rates and are not affected by this wagon-wheel effect can explain the recovery of detection performance for modulation rates of 100 Hz and higher. If we assume that for the detection of echo-amplitude modulation bats use amplitude differences between consecutive echoes, we should observe a similar wagon-wheel effect for the current sensitivity for amplitude modulation. However, we did not observe a drop in performance at intermediate amplitude-modulation rates (Fig. 3). This indicates that the currently observed increasing sensitivity for increasing amplitude-modulation rates does not purely reflect the ability to detect echo-amplitude differences. It rather indicates that the bats employ a different detection strategy already at low modulation rates where the wagon-wheel effect does not yet occur.
We propose a detection strategy that depends on spectral cues rather than amplitude cues. We simulated the interaction of a frequency-modulated echolocation call with a fast amplitude modulation and observed that in an amplitude-modulated echo, different frequency bands get emphasized and suppressed in comparison to an unmodulated echo (Fig. 5). We quantified these changes by calculating the weighted mean of the frequencies present in the echo, i.e., the echo spectral centroid. The variation in echo amplitude between different phases of the amplitude modulation decreased with increasing modulation rates (Fig. 6, black line). In direct contrast to this, variation in spectral centroid between different phases of the amplitude modulation increased with increasing modulation rates (Fig. 6, gray line). In other words, amplitude cues become less and spectral cues become more available with increasing modulation rate. The bats’ behavioral performance in detecting amplitude modulations can therefore be explained by a shift in the emphasis that they place on processing different auditory cues. We can only guess at the details of the perceptual weighting that the bats might apply towards amplitude and spectral (and possibly further) cues, however, the high-pass shape of the behavioral modulation transfer function suggests a strong contribution of the spectral cues.
The idea that the perception of echo-amplitude modulation is processed differently from the perception of echo-delay modulation is corroborated by observations during the training procedure: all four bats that participated in the current amplitude-modulation experiments had previously participated in the delay-modulation experiments. The rewarded stimulus in these two experiments was the same unmodulated reflection. The bats were thus familiar with virtual targets, the experimental set-up and procedure. However, while these bats needed very little retraining when we wanted to acquire data for a new delay modulation rate, the same individuals needed extensive retraining periods when we wanted to acquire data for a new amplitude modulation rate. Retraining periods are illustrated in Fig. 7. The data clearly show that on average the animals needed more than five times the retraining time when we changed the amplitude-modulation rate than when we changed the delay-modulation rate. These data, together with the difference in psychophysical performances suggest that the modulation of echo amplitude may be processed fundamentally differently from the modulation of echo delay. We hypothesize that the dedicated neural circuitry for echo delay, and specifically the topographic representation of echo delay in the bat auditory cortex, supports delay-modulation detection (Hagemann et al. 2010; O’Neill and Suga 1979; Dear et al. 1993), whereas the lack of a similar topographic representation for echo amplitude may underlie the much more demanding training and data acquisition for amplitude-modulation detection. Modulation at high rates induces spectral cues for both echo-delay modulation and echo-amplitude modulation. These spectral cues are readily represented along the tonotopic axes at virtually all stages of the bat auditory system. The fact that our bats learned to detect fast delay modulations much quicker than fast amplitude modulations may be related to the fact that the delay-induced Doppler distortions create overall much more dramatic spectral distortions than the above described spectral changes induced by the echo-amplitude modulation.
Comparison to passive listening
In passive listening, studies that investigate sensitivity to amplitude modulation typically use broadband noise as the carrier signal for the modulation, as opposed to a sequence of echoes of self-produced, broadband echolocation calls. The sensitivity of mammals and birds to amplitude modulation of noise consistently becomes worse with increasing modulation rate (Viemeister 1979; Salvi et al. 1982; Burdin et al. 1973; Dooling and Searcy 1981).
These findings are in direct contrast to the results we present here for the sensitivity of bats to amplitude modulation of echoes, and indicate separate processing mechanisms for amplitude modulations in passive and active listening, at least for part of the range of modulation rates. This discrepancy can be assigned to the fundamental difference in the nature of the modulated sounds. In passive listening, the carrier of the amplitude modulation is a continuous signal that persistently reflects every point of the modulation. In active listening, the carrier of the amplitude modulation is a transient, i.e. impermanent signal that reflects only momentary stages of the modulation.
When the carrier of the amplitude modulation in passive listening is a pure tone, there also arise spectral cues that allow the detection sensitivity of human listeners to recover for modulation rates higher than 50 Hz (Viemeister 1979). Fast amplitude modulation of a narrowband signal creates spectrally resolvable sidebands: when a 1000 Hz pure tone is amplitude-modulated at a modulation rate of 200 Hz, the sidebands are at 800 and 1200 Hz. If we regarded the frequency-modulated broadband echolocation call as an assembly of pure-tone carriers, the broadband echo itself would mask any emerging frequency sidebands, even at the highest presented amplitude modulation rate of 500 Hz. Instead, the observed changes in the frequency spectrum of amplitude-modulated echoes (Figs. 5, 6) are likely to serve the same purpose as frequency sidebands in passive listening, complementing the detection of echo-amplitude modulation for those (high) modulation rates where this strategy is more efficient than resolving temporal amplitude differences.
Dankiewicz et al. (2002) performed the only other perceptual study on amplitude modulation detection with echolocation, i.e. not in passive listening. They trained a bottlenose dolphin (Tursiops truncatus) to discriminate modulated synthetic echoes from unmodulated ones. In contrast to the current work, the authors did not generate real-time echoes from the animal’s emissions but only used the animal’s emission to trigger the playback of the synthesized echo. Second, the authors applied one amplitude modulation cycle across a fixed number of echoes (8 to 64), not across a fixed time; thus the effective modulation frequency depended on the number of the dolphin’s emissions per time. Modulation detection thresholds changed from about 1.2% (= − 38 dB) for an effective modulation frequency around 2 Hz to a threshold of about 6% (= − 24 dB) for an effective modulation frequency around 16 Hz. Thus the modulation transfer function was low-pass, similar to what has been reported for amplitude-modulation detection with wideband noise carriers in humans (Viemeister 1979). Although the authors did not test the dolphin at effective modulation rates higher than 16 Hz, the general potential for spectral echo changes is low: interactions between the amplitude modulation of the virtual target and the frequency modulation of the emitted call are limited in the dolphin, because its average echo duration is markedly shorter than in our bats [128 µs and 500 µs, respectively (Dankiewicz et al. 2002)]. Instead, with regard to perceiving amplitude differences between consecutive echoes at very small modulation depths, echolocating dolphins are presumably superior to bats, as they have repeatedly been reported to be capable of detecting a 1 dB difference in target strength (Evans 1973; Bullock et al. 1968; Johnson 1967; Moore et al. 1995) compared to 5–7 dB in P. discolor (Heinrich et al. 2011). Additionally, dolphins may have no need to detect modulation rates that would be high enough to create spectral cues. While echo-amplitude modulation can reflect the changing orientation of the target relative to the emitter both in bats and dolphins, the speed of these orientation changes is likely much slower underwater than in air. All of these aspects make dolphins more likely to solely use echo-amplitude cues, resulting in their reported low-pass modulation transfer function.
As opposed to dolphins, FM bats would greatly benefit from a mechanism to facilitate flutter detection at high modulation rates, because these make up a large portion of insect prey wing beat rates (Pringle 1949; Sotavalta 1953; Gibson et al. 2010). Bats from the Eocene radiation were nocturnal and insectivore (Speakman 2001; Veselka et al. 2010). The surprisingly good performance of our bats (especially at higher modulation rates) leads us to hypothesize that the biophysical properties of FM echolocation calls do indeed facilitate insect flutter detection, not only based on echo-delay modulations, but also based on echo-amplitude modulations. We suggest that the frequency-modulated call structure does not only reflect a motor constraint of laryngeal echolocation or a means to minimize Doppler-distortion-induced misjudgments in echo delay (Altes 1995; Simmons et al. 2004). We argue that adaptive selection to insect prey motion shaped FM echolocation calls not into a perfect Dirac Impulse but into a structure that converts fast amplitude modulations into spectral cues, which are readily represented along the bats’ tonotopic axis. We surmise that the small deviation from the Dirac structure grants FM bats sensitivity to fast-changing time-variant environments together with high spatial acuity in time-invariant scenarios.
In summary, our work offers insights into the processing of modulated echo amplitude by FM bats. We have introduced a virtual-reality approach with time-variant targets to assess sensitivity to echo-amplitude modulation independently of echo-delay modulation. We have shown that FM bats are well capable of detecting modulations of echo amplitude despite the limitations that arise from the use of very short echolocation signals. We suggest that amplitude-modulation detection with echolocation not only differs fundamentally from delay-modulation detection, but also from amplitude-modulation detection in passive listening, due to the transient nature of the carrier signal. We speculate that the mechanism to detect (particularly the fast) modulations does not rely on the nominal cue alone, the echo amplitude, but on spectral cues that occur when the frequency modulation of the echolocation call interacts with the amplitude modulation of the target. Although we do not yet know whether FM bats make use of echo-amplitude modulations when they encounter a moving target, we provide an important proof-of-principle demonstration that offers release from the supposed trade-off between temporal and spatial acuity for FM bats.