Bats Use Echo Harmonic Structure to Distinguish Their Targets from Background Clutter

See allHide authors and affiliations

Science  29 Jul 2011:
Vol. 333, Issue 6042, pp. 627-630
DOI: 10.1126/science.1202065


When echolocating big brown bats fly in complex surroundings, echoes arriving from irrelevant objects (clutter) located to the sides of their sonar beam can mask perception of relevant objects located to the front (targets), causing “blind spots.” Because the second harmonic is beamed more weakly to the sides than the first harmonic, these clutter echoes have a weaker second harmonic. In psychophysical experiments, we found that electronically misaligning first and second harmonics in echoes (to mimic the misalignment of corresponding neural responses to harmonics in clutter echoes) disrupts the bat’s echo-delay perception but also prevents clutter masking. Electronically offsetting harmonics to realign their neural responses restores delay perception but also clutter interference. Thus, bats exploit harmonics to distinguish clutter echoes from target echoes, sacrificing delay acuity to suppress masking.

Echolocating big brown bats (Eptesicus fuscus) broadcast brief frequency-modulated (FM) sounds with two prominent acoustic harmonics—FM1 and FM2 (Fig. 1A) (1, 2). The principal perceptual dimension in bat sonar is target range, as determined from echo delay (3). Although these bats typically catch insects in the open, they also can capture prey near vegetation and other obstacles, even while flying in groups (4, 5). Bats aim their sonar forward to track targets and fly through vegetation, but the beam is wide enough to impinge on surrounding objects, called clutter (Fig. 1B). Echoes from clutter return to the bat at variable delays (6, 7), and they can mask or degrade echoes from the target of interest. Here we investigate how the bat is able, while flying effectively in complex surroundings (8, 9), to segregate the echoes of relevant targets from those of irrelevant clutter.

Fig. 1

(A) FM biosonar sound with harmonics FM1 and FM2. (B) Wide broadcast beam (pink sector) impinges on target in front as well as vegetation to side. Overlapping echoes form clutter zone (green shading) with “blind spots” that mask the target. (C) Horizontal beamwidth is narrower for FM2 (60 kHz, blue; 80 kHz, tan) than FM1 (25 kHz, red; 40 kHz, green) (10, 11). (D) Broadcast amplitude declines as 1/distance due to spreading (gray), but atmospheric absorption is stronger for FM2 than for FM1 (14). (E) Color-density contour plot showing amplitude of FM2 relative to FM1 (0 to −48 dB) according to direction and distance.

The horizontal width of the sonar beam in big brown bats is dependent on frequency, ranging from ±70° at 25 kHz in FM1 to ±30° at 80 kHz in FM2 (Fig. 1C) (10, 11). When the bat tracks a target (with ±1° to 5° accuracy) (12, 13), it receives the full sonar broadcast. Sounds that impinge on objects located off the central beam’s axis are lowpass filtered—that is, they are weakened more at the high frequencies in FM2 than at the low frequencies in FM1 (Fig. 1C). Echoes from more distant, nontarget objects are reduced in overall amplitude but also lowpass filtered (FM2 < FM1) by atmospheric attenuation (Fig. 1D) (14). Thus, FM2 is weaker than FM1 in echoes from objects located off the central axis or at greater distances. Echoes from the target are derived from interference between multiple, overlapping reflections that reinforce and cancel each other at specific frequencies, not from global lowpass filtering as is the case with nontarget echoes (3, 15, 16). Therefore, the ratio of FM2 to FM1 in sounds impinging on specific locations defines the harmonic broadcast beam (Fig. 1E). For example, echoes arriving from a nearby target zone about ±20° wide have a “flat” spectrum (FM2 = FM1), whereas echoes arriving from the sides or farther away have lowpass spectra (FM2 < FM1). We hypothesize that big brown bats exploit the relative weakness of FM2 in echoes to categorize clutter as irrelevant and mitigate masking.

To evaluate clutter masking (blind spot in Fig. 1B), we used a two-choice experiment for discrimination between echoes with different delays (Materials and Methods are available as supporting material on Science Online.) Trained bats sat on an elevated Y-shaped platform and emitted sonar sounds (Fig. 2A). The bat’s broadcasts (Fig. 2B) were picked up by microphones (m), delayed electronically, and delivered back to the bat from loudspeakers (s) at the ends of each arm of the platform to simulate echoes of targets (17). Bats were trained with food reward to respond to the test echo (the “target,” or positive stimulus, S+) by moving forward toward the loudspeaker on the corresponding arm of the platform. For each trial, a second, probe echo (the unrewarded “clutter,” S−) was presented from the other loudspeaker to assess clutter masking. The test echo, S+, was a two-glint (a glint is a single reflecting point) simulated target with glint delays of 3160 and 3460 μs (300-μs glint separation; Fig. 2C). These two delays corresponded to target ranges of 55 and 60 cm. The probe echo, S−, was a one-glint simulated object (Fig. 2, D to F). It was presented at a series of different delays (in steps from 3660 down to 3060 μs) covering a sufficiently wide span to trace the size of the blind spot it induces for each of the two glints in the test echo. [Echo strength for each glint was −35 dB rebroadcast strength (17)]. Data are from four big brown bats; 150 trials per bat per value of S− delay. We defined perfect performance as 0% discrimination errors; typical baseline performance was about 5% errors, and chance was at 50% errors. A marked increase in discrimination errors above baseline performance occurred when probe echoes fell within ±50 μs of either glint in test echoes, thus indicating masking (18, 19). Masking also depends on the angular separation between test and probe echo sources. Even when they are separated by as much as 40°, masking still occurs if echo delays are similar (20). Figure 3A shows a separate error peak ±50 μs around the delay of each glint in S+ (gray curve for flat-spectrum clutter S− in Fig. 2D).

Fig. 2

(A) Two-choice behavioral experiment used to measure masking for clutter interference. Bat is rewarded for moving toward S+ echoes (black arrow). (B) Broadcast sound containing FM1 and FM2. (C) Two-glint S+ echo with multiple interference notches at 3.3-kHz intervals due to 300-μs separation of glints. (D) One-glint flat-spectrum S− echo used to demonstrate masking of S+ [inset graph in (A) shows masking effect as error peaks]. (E) Lowpass-filtered S− with FM2 weaker than FM1 by 1.5 dB. (F) Lowpass-filtered S− with FM2 at earlier delays relative to FM1 to compensate for amplitude-latency trading.

Fig. 3

(A) Normal masking (gray circles) shows as discrete error peaks when delay of flat-spectrum S− echoes (Fig. 2D) matches the delays of two-glint S+ echoes (Fig. 2C). Masking persists with electronic “split-harmonic” filters that separate FM2 from FM1 in a narrow band around 55 kHz (green triangles). Lowpass filtering of S− (Fig. 2E) abolishes masking (pink triangles). When S+ is split-harmonic echo with FM2 delayed 300 μs after FM1, masking still is present (red diamonds), but error peaks widen and are surrounded by wide plateau of errors, illustrating defocused delay image. (B) Expanded plot of masking by flat-spectrum S− (green triangles) and no masking by lowpass S− (pink triangles) from (A). When delay of FM2 is advanced by 0 to 50 μs relative to FM1 to counteract amplitude-latency trading (~24 μs) caused by 1.5-dB lowpass attenuation of FM2, masking is restored for glint 1 (dark blue circles) but not for glint 2. (C) Defocusing effect shown in (A) is activated by FM2 delays as small as 3 μs (red diamonds compared to yellow diamonds). Masking recovery data from (A) are replotted (light blue circles) and then shifted rightward by 24 μs (dark blue circles) to compensate for amplitude-latency trading. This is to align FM2 to the same delay of FM1 (0 μs) as in the defocusing experiment. Defocusing effect builds up for FM2-FM1 differences from 0 to 5 μs, while masking declines in reciprocal fashion. All plots show mean percent errors ±1 SD from binomial distribution.

To test the hypothesis that bats detect the weaker FM2 in lowpass echoes to recognize clutter, we introduced electronic “split-harmonic” filters to separate FM1 from FM2 in S− for individual manipulation without changing FM1 and FM2 otherwise (17, 21) (Fig. 2D). With these split-harmonic filters, a blind spot for masking still occurs around ±50 μs of each glint (Fig. 3A, green curve). Then, to determine whether weaker FM2 is recognized and used to suppress masking, we mildly lowpass filtered (FM2 < FM1 in Fig. 2E) the clutter echo (S−) to achieve a mean FM2 attenuation of 1.5 dB. This slight lowpass filtering mimics echoes arriving from nearby clutter located ~15° off-axis (Fig. 1, C and E). The bats’ performance shows no error peak—no blind spot—for either glint in S+ (Fig. 3A; purple curve). Slightly weakening FM2 in S− thus abolishes its masking effect on S+ (green versus purple curves on expanded plot in Fig. 3B).

The key to how the bat recognizes weaker FM2 from lowpass filtering is an interaction between the amplitude of an echo and the timing, or latency, of the neural responses it evokes—an effect called amplitude-latency trading (22). In big brown bats, attenuation of echoes retards echo-evoked neural responses in the auditory midbrain by 15 to 16 μs per dB (19, 2325). Prolonging response latency in turn directly affects the bat’s perception—increasing the perceived delay by 15 to 16 μs/dB (19, 25). Reducing the strength of FM2 relative to FM1 (i.e., lowpass filtering) thus retards responses to FM2, which misaligns the timing of responses to the harmonics; however, does harmonic misalignment affect the bat’s perception of delay? Deliberately introducing electronic misalignment so that FM2 follows FM1 by 300 μs has been found to cause a marked loss in the bat’s echo-delay acuity. The error peaks for each glint in S+ increase in width from ±50 μs (gray or green curves in Fig. 3A) to cover a total span of about 1000 μs (red curve in Fig. 3A), so that the two glints blur together and become merged in perception (17). In effect, the bat loses so much acuity for perceiving the delay of S+ that its delay image is defocused. Misalignment of FM2 after FM1 by as little as 3 μs has been found sufficient to initiate defocusing, which reaches its maximum for 5 to 25 μs (Fig. 3C, red curve compared with yellow curve for control) (21). Defocusing depends specifically on temporal misalignment of neural responses to FM2 in relation to FM1. For example, reducing FM2 by 3 dB relative to FM1 causes neural responses to be retarded by about 48 μs relative to FM1 (24, 25), enough to cause defocusing (21). Shifting FM2 forward in time by 48 μs compensates for amplitude-latency trading and restores focused delay acuity (26).

Is defocused delay acuity associated with masking? In the lowpass S− echoes (Fig. 2E) that mimic clutter located about 15° off to the side (Fig. 1E), attenuation of FM2 relative to FM1 averages 1.5 dB. Amplitude-latency trading should retard neural responses to FM2 ~24 μs longer than responses to FM1 (24, 25), which is enough to cause full defocusing (red curve, Fig. 3C) (21). We presented the bats with S− at a delay of 3160 μs (which masks glint 1; green curve in Fig. 3B) or 3460 μs (which masks glint 2; green curve in Fig. 3B), applied lowpass filtering to remove masking (purple curve in Fig. 3B), and then compensated for amplitude-latency trading by sliding FM2 to earlier delays of 0 to 50 μs relative to FM1 (see Fig. 2F) while keeping FM1 at one or the other fixed delay. If defocusing is removable by adjusting the harmonics to realign their neural responses (21, 26), masking should be restored when FM2 leads FM1 by about 24 μs. Masking indeed returns (i.e., an error peak again appears) for glint 1 in S+ when FM2 is advanced by 25 to 45 μs (Fig. 3B, dark blue curve at glint 1). Masking does not recover for glint 2 in S+ (Fig. 3B, dark blue curve at glint 2). This distinction may exist because the 300-μs delay of glint 2 relative to glint 1 is not represented by the timing of neural responses evoked directly by glint 2 echoes but instead by the series of evenly spaced notches in the two-glint spectrum at 3.3-kHz intervals (Fig. 2C) (19).

Figure 3C compares defocusing (red curve) (21) with the advancement of FM2, introduced here to compensate for amplitude-latency trading and restore masking for glint 1 (light blue curve). For direct comparison, the relative timing of neural responses to the harmonics in the masking experiment needs to be realigned to match the relative timing of responses to harmonics in the defocusing experiment. To do this, the masking recovery results (Fig. 3C, light blue curve) are shifted 24-μs rightward (Fig. 3C, dark blue curve). Masking of S+ by S− occurs as a trade-off in relation to defocusing of the delay image for S+ (Fig. 3, dark blue versus red curves).

To prevent masking of echoes from objects located immediately to the front, big brown bats exploit the harmonic structure of their broadcast beam (Fig. 1E) to impose categorical perception on “clutter,” which is defocused, as distinct from “target,” which is focused. Although flat-spectrum echoes can mask perception of target glints (Fig. 3A, gray curve; Fig. 3AB, green curve), clutter echoes are lowpass filtered by their location in the harmonic beam and rendered both defocused and not capable of masking (Fig. 3C). The bat suppresses interference by sacrificing delay acuity for echoes from clutter while retaining acuity for echoes from targets.

Supporting Online Material

Materials and Methods

Figs. S1 to S4

References and Notes

  1. Acknowledgments: 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 to J.A.S.; from a Brown University dissertation fellowship to M.E.B.; from ONR-Global grant N00014-07-1-0857 to T.V.Z.; and from the RI Space Grant and from the Brown Institute for Brain Science. We thank A. M. Simmons for editorial advice.
View Abstract

Navigate This Article