Bats use a neuronally implemented computational acoustic model to form sonar images
Graphical abstract
Highlights
► Neural processing in FM bat sonar is examined to find a level at which echolocation-specific information is manifested, not just responses to acoustic features. ► The occurrence of only one spike per sound highlights the possible temporal neural basis for perception. ► Neural processing combines echo delay and the echo spectrum into a single perceptual image. ► Parallel neuronal representations persist at all levels of processing. ► The progression of neuronal responses over latency resembles an internal computational model of sonar acoustics.
Introduction
The biological sonar of bats is a core model system for a neuroethological approach to neuroscience [1]. Much has been learned about echolocation sounds, echolocation behavior, and the auditory basis for echolocation—all in the context of bat biology in general [2•]. In support of this approach, there are many studies of neural responses in the auditory systems of echolocating bats to such standard stimuli as tone-bursts as well as stimuli that mimic echolocation broadcasts and echoes [1, 3]. Experiments have assessed which parameters of sounds have the greatest effect on single-unit responses in different auditory structures, what patterns of spikes are evoked, whether there are binaural interactions, and how inhibitory as well as excitatory inputs govern the occurrence of spikes. At the single-cell level, responses are typically characterized by the profile of stimulus features that most affect their occurrence [3]. Viewed from these features, auditory responses in echolocating bats are surprisingly similar to auditory responses recorded from other, nonecholocating mammals [3, 4]. Aside from activation by ultrasonic frequencies, the responses could just as well have been recorded from the auditory systems of numerous species of mammals, birds, or amphibians [5•, 6•, 7]. One possibility, of course, is that bats have co-opted normal auditory processes, such as those for pitch perception, to serve their own special purposes. For example, the time delay of echoes associated with a train of sonar broadcasts is analogous in some respects to the periodicity of complex sounds [5•, 6•], which could lead to common neuronal response properties serving equally well for hearing and echolocation [7]. Still, echolocation is a dramatic adaptation of hearing for spatial perception, and there ought to be some level at which neuronal responses in bats manifest its existence.
Most species of bats emit frequency-modulated (FM) echolocation sounds that cover a substantial span of frequencies and use echoes of these sounds for spatial sensing of the immediate environment [1]. As a broad principle, sonar systems that transmit wideband signals take advantage of the many frequencies in their broadcasts to enhance acuity for determining distance from echo delay and to classify objects from echo spectra [8•]. However, sonar receivers often do more than just make measurements—they can incorporate knowledge about the environment into an internal computational representation of the environment's structure—a practice called model-based signal processing [9]. This review offers the hypothesis that auditory responses in echolocating bats can be viewed as embodying an internal model of sonar acoustics and echo formation. To produce wideband sonar images, information has to be integrated across the many frequencies in the sounds [8•], but single neurons at all levels of the bat's auditory system are predominantly excited by a single frequency region around which inhibitory interactions frequently occur [1, 3]. No neurons have been found in bats that exhibit the integrative characteristics in their responses that would be required to act as a unitary ‘display’ of objects. If individual cells do not register information that bridges across frequency regions in the sounds, it may be because the critical integrative processes in echolocation are not organized at the single-cell level. This review steps back from the intrinsically atomistic view of response properties occurring in single cells to identify a level of functional neuronal description where the fact of echolocation is more evident. At this level, in terms of model-based signal processing, the bat's auditory system can be viewed as operating an internal, neuronally executed computational reconstruction of the acoustic environment for sonar.
Section snippets
Echolocation by big brown bats
Big brown bats (Eptesicus fuscus) emit trains of wideband (20–100 kHz) FM sounds and perceive objects in the surrounding scene by listening for echoes that return to their ears [10••]. These bats use their sonar to hunt for flying insects and to guide flight in complex surroundings such as vegetation [1, 2•, 10••]. Figure 1a shows a big brown bat flying in a darkened room filled with rows of plastic chains hanging from the ceiling [11]. The spectrogram of a typical broadcast is illustrated in
Auditory responses to FM sounds
When a big brown bat transmits an FM sonar signal, it picks up that sound with its inner ears and encodes it as auditory responses. Afterwards, when it receives one or more echoes of that sound (Figure 1b), these, too, are picked up by the inner ears and encoded. The ensuing neural activity combines signal encoding [1, 3], auditory regulation [30], sensorimotor control [31, 32], and image formation [33]. Although, at the single-cell level, auditory processing is usually described in terms of
The bat's internal model of broadcast-echo propagation
Figure 3 shows plots of spike patterns recorded from single neurons in the big brown bat's IC and auditory cortex (AC) [34, 35, 36•, 37]. In the IC, the principal response properties are the cell's BF and response latency (Figure 3a–e). Individual neurons are tuned to their BF with varying sharpness of tuning (Figure 3a)—and have a stable on-response latency at BF (Figure 3e). Crucially, in different cells tuned to the same BF, latencies of on-responses are dispersed over a span from 4–5 ms to
Responses specific to echoes
While the broadcast sound propagates acoustically outward from the bat to impinge on objects up to 5–8 m away and return as echoes at delays up to 30–50 ms, the time-frequency replica of the broadcast propagates neuronally through the IC for up to 30–50 ms (Figure 3e). When an echo returns, it evokes its own auditory spectrogram at the inner ear and cochlear nucleus. Spikes representing successive frequencies in the newly received echo eventually reach the IC, where they begin a new time-frequency
The bat's internal model of target acoustics
The physiological effects of having nulls in the echo spectrum (Figure 3c,d; Figure 4a,b) lead to consideration of the bat's internal model for target structure as a component of its hypothetical model for broadcast-echo propagation. By ensonifying objects over a wide frequency band (Figure 1a), the bat gathers information about target structure to perceive target shape using distinctive features of the spectrum of echoes relative to broadcasts [24]. Flying insects [27]—or the artificial
Conclusion
This review presents the hypothesis that, when the bat emits sonar sounds, it picks up each broadcast with its ears and initiates operation of a dynamic, real-time internal auditory model of the sound's radiation away from the bat—a model implemented by neuronal responses distributed in time for 30–50 ms across large numbers of IC neurons tuned to different frequencies. The read-out for the model's propagation process, in the AC, explicitly includes sensitivity to local spectral nulls as cues
References and recommended reading
Papers of particular interest, published within the period of review, have been highlighted as:
• of special interest
•• of outstanding interest
Acknowledgements
Support for this research came from Office of Naval Research (ONR) grants N00014-04-1-0415 and N00014-09-1-0691, NSF grant IOS-0843522, and National Institute of Mental Health grant R01-MH069633. The author is grateful for the extensive contributions from members of the Brown Bat Lab, and wishes to thank John R. Buck, of UMass Dartmouth, and James V. Candy, of the Lawrence-Livermore National Laboratory and UC Santa Barbara, for bringing model-based signal-processing to his attention.
References (45)
Blip, Ping, and Buzz: Making Sense of Radar and Sonar
(2007)- et al.
Frequency tuning and response latencies at three levels in the brainstem of the echolocating bat, Eptesicus fuscus
J Comp Physiol A
(1994) The Biology of Bats
(2000)Bats: From Biology to Conservation
(2011)- et al.
The Neural Basis of Echolocation in Bats
(1989) - et al.
Bats and frogs and animals in between: evidence for a common central timing mechanism to extract periodicity pitch
J Comp Physiol A
(2010) On the role of space and time in auditory processing
Trends Cogn Sci
(2001)- et al.
A neuroethological theory of the operation of the inferior colliculus
Brain Behav Evol
(1996) Bayesian Signal Processing: Classical, Modern, and Particle Filtering Methods
(2009)