Discrimination of jittered sonar echoes by the ... · ic maps in the bat's auditory system presumably provide a topographic display of echo spectral features-in par- ticular, the

J Comp Physiol A (1990) 167: 589-616 Jmmml of S~mtor/,

and

P h y s i o l o g y A ~-~= physiology

�9 Springer-Verlag 1990

Discrimination of jittered sonar echoes by the echolocating bat, Eptesicus fuscus : The shape of target images in echolocation

James A. Simmons 1, Michael Ferragamo 2, Cynthia F. Moss*, Scott B. Stevenson**, and Richard A. Altes 3

1.2 Department of Psychology and Section of Neurobiology, Division of Biology and Medicine, Brown University, Providence, RI 02912, USA

3 Chirp Corporation, 8248 Sugarman Drive, La Jolla, CA 92037, USA

Accepted August 18, 1990

Summary. 1. Behavioral experiments with jittering echoes examined acoustic images of sonar targets in the echolocating bat, Eptesicus fuscus, along the echo delay or target range axis. Echo phase, amplitude, bandwidth, and signal-to-noise ratio were manipulated to assess the underlying auditory processes for image formation.

2. Fine delay acuity is about 10 ns. Calibration and control procedures indicate that this represents temporal acuity rather than spectral discrimination. Jitter discrimination curves change in phase when the phase of one jittering echo is shifted by 180 ~ relative to the other, showing that echo phase is involved in delay estimation. At an echo detectability index of about 36 dB, fine acuity is 40 ns, which is approximately as predicted for the delay accuracy of an ideal receiver.

3. Compound performance curves for 0 ~ and 180 ~ phase conditions match the crosscorrelation function of the echoes. The locations of both 0 ~ and 180 ~ phase peaks in the performance curves shift along the time axis by an amount that matches neural amplitude-latency trading in Eptesicus, confirming a temporal basis for jitter discrimination.

Key words: Echolocation - Biosonar - Auditory system - Acoustic images - Neural timing - Target ranging

Introduction

Many species of echolocating bats emit frequency- modulated (FM) sonar sounds and perceive objects from echoes returning to their ears (Busnel and Fish 1980; Griffin 1958; Nachtigall and Moore 1988). In echolocation, the bat's ears and auditory system function as the

* Presen t address: Department of Psychology, Harvard University, Cambridge, MA 02138, USA ** Present address: School of Optometry, University of California, Berkeley, Berkeley, CA 94720, USA

sonar receiver and transform the acoustic waveform of echoes into images having spatial dimensions (Simmons et al. 1990). In essence, the echolocation of bats is a sophisticated real-time acoustic imaging system that serves these animals for orientation in the environment, particularly for finding prey and avoiding obstacles to flight. This paper reports experiments that confirm and extend earlier observations of a crucial relationship between the waveform of signals and the shape of images perceived by the echolocating bat, Eptesicusfuscus (Sim- mons 1979). This relationship reveals fundamental characteristics of the signal-processing operations used by the bat's sonar receiver.

Target ranging

One of the primary functions of echolocation is perception of the distance to objects, or target range. Bats determine distance from the time required for their FM echolocation signals to travel outward to the target and then return to the ears (Simmons 1973). Each centimeter of target range adds about 58 las to the delay of echoes. For insect-sized targets, the practical operating range of echolocation is restricted by atmospheric attenuation and spreading losses to about 3 to 5 m (Griffin 1958; Kick 1982), resulting in echo delays of as much as 17 to 30 ms. The approach and tracking stages of pursuit (Kick and Simmons 1984), during which the bat carries out most of its evaluation of targets, cover distances of about 20 to 150 cm, with associated echo delays from roughly 1 to 8 ms. Perception of the absolute range of objects corresponds to perception of echo delay, which is directly encoded by the timing of neural discharges evoked by FM sweeps in echoes (Bodenhamer and Pollak 1981; Bodenhamer et al. 1979; Pollak et al. 1977; Suga 1970) and then displayed topographically by the bat's auditory nervous system through neural responses tuned to particular ranges (O'Neill and Suga 1982; Suga 1988; Suga

590 J.A. Simmons.et al.: Jittered sonar echoes in echolocation

and Horikawa 1986; Suga and O'Neill 1979; Sullivan 1982; Wong and Shannon 1988). The bat hears each emission, marking the occurrence of each emitted frequency with discharges that are transmitted upward through the auditory pathways of the brain. Discharges evoked a short time later by corresponding frequencies in echoes also ascend the auditory pathways, and a neural network (see Licklider 1951 ; Lazzaro and Mead 1989) uses coincidences of emission discharges and echo discharges to transform the initial temporal code for echo delay into a place code (Kawasaki et al. 1988; Suga 1988).

The process of target ranging is complicated by the convergence of information about the shape of a target onto the same psychological scale used to perceive distance. Complex targets can be described in terms of a series of reflecting points, called 9lints, located at different ranges (Altes 1976). The range separation of the glints, or the target's range profile, is the equivalent of shape to Eptesicus. If the time separation of the echo components from different glints is smaller than about 300 to 350 gs (the peripherally-determined integration- time for echo reception), only the earliest component evokes delay-coding neural discharges that can directly register target range (Simmons et al. 1989). Because the sonar signals themselves are several milliseconds long, subsequent echo components largely overlap with this first component to create an interference spectrum for the compound echo as a whole (Altes 1984; Beuter 1980). The distribution of peaks and notches across the spectrum represents the time-separation of the components, and neural coding of this spectral shape makes up for the absence of timed neural discharges to echo components following within the integration-time window. Tonotop- ic maps in the bat's auditory system presumably provide a topographic display of echo spectral features-in particular, the distribution of peaks and notches across different frequencies.

Eptesicus and other species of FM bats can discriminate between targets having different range profiles and thus different echo spectra (Habersetzer and Vogler 1983; Schmidt 1988; Simmons et al. 1974). Although information about the range profile of targets initially is encoded in terms of echo spectra (see Schmidt 1988), it appears ultimately to be expressed perceptually in terms of range itself, along the same psychological axis form- ally calibrated by the timing of neural discharges to echoes (Simmons et al. 1990; Simmons and Stein 1980). In particular, the spectral representation of target shape is transformed into its equivalent time-domain representation for incorporation into the perceived image. Thus, the axis of range is a synthetic or computed dimension of perception that incorporates information initially encoded both in time and in frequency. This duplex character makes perceived range analogous to the pitch of sounds, which also is a product of time and frequency representations (Licklider 1951; Wever 1949; see Laz- zaro and Mead 1989). Information that is separately displayed on topographic neural maps representing echo delay and echo spectra must be combined to create this computed range axis. Furthermore, any understanding

of target ranging must take into account the dual temporal/spectral character of the neural representations which can write information onto the range axis. The research described here seeks to identify the signal- processing operations that transform echo waveforms into acoustic images in a manner that preserves both temporal and spectral cues for range. This is the fourth paper in a series that explores the mechanisms of acoustic imaging by FM bats (Simmons and Chen 1989; Simmons et al. 1989, 1990).

SignaLprocessin 9 for target range

In a theoretical sense, target ranging is a fundamental aspect of the performance of sonar systems, and much can be discerned about the nature of the sonar receiver from the accuracy of registration of images of targets along the axis of delay or range (Woodward 1964; Van Trees 1971). Considerable research on echolocation in bats has focused on target ranging ever since early spe- culations that the FM echolocation signals of some species of bats might be suited for processing by a pulse- compression or crosscorrelation type of sonar receiver (Strother 1961; van Bergeijk 1964; McCue 1966). A pulse-compression receiver replaces each sonar echo with the crosscorrelation function between the sonar emission and the echo. The characteristics of emissions and echoes then are judged in terms of their associated crosscorrelation functions, and the design of a good sonar signal becomes a matter of tailoring the transmitted waveform to have a good crosscorrelation function. Because it is intended to mark the echo's time-of-arrival as sharply and unambiguously as possible, a good crosscorrelation function will have a single central peak with as few additional peaks of substantial amplitude as possible located nearby to be confused with the central peak. That is, side peaks should be low in amplitude and remote from the central peak. The location of the central peak is the best estimate of echo delay, and prominent side-peaks could be confused with this estimate, reducing the accuracy of target ranging. This definition of a good crosscorrelation function applies to more complex perceptual tasks than just target ranging. For example, when multiple targets are present as clutter interference, prominent side-peaks in the crosscorrelation function for echoes from clutter could mask the location of the main peak for the target of interest, thus widening undesirably the effective size of the clutter.

Three acoustic characteristics of sonar transmissions determine the shape of the crosscorrelation function for echoes (Simmons and Stein 1980). If the sonar signal is broad in bandwidth (centralized RMS bandwidth; Menne and Hackbarth 1986), the envelope of the crosscorrelation function as a whole is narrow in time, and it accordingly marks the moment of echo reception very sharply. Under the envelope, the function has a fine structure that reflects details of the signal's frequency composition. A broad signal bandwidth provides a narrow envelope that is very restrictive to the fine structure of the crosscorrelation function lying underneath, which

J.A. Simmons et al. : Jittered sonar echoes in echolocation 591

is desirable because there will be fewer side-peaks to be confused with the central peak. This fine structure depends upon two additional factors - the average frequency (RMS bandwidth; Menne and Hackbarth 1986) and the average first-harmonic period of the signal if it has a harmonic organization (Simmons and Stein 1980). The waves making up the fine structure have shapes that correspond to cycles of the average frequency, and their width is the reciprocal of this average frequency. The higher the average frequency, the narrower are the individual peaks in the fine structure. The heights of these peaks are restricted by the crosscorrelation function's envelope, which depends only upon bandwidth, not the absolute values of the frequencies in the signal. The central peak of the fine structure can be made desirably narrow by keeping the frequencies in the sonar signal relatively high, while confusing side-peaks can be reduced in amplitude by having a narrow envelope lying over the fine structure. These side-peaks are trapped underneath the bandwidth-determined envelope, and they become lower in amplitude on either side of the central peak if the signal's bandwidth is raised.

Having harmonic structure in the signal provides a further means for reducing the height of side-peaks in the fine structure of the crosscorrelation function. The spacing between adjacent peaks in the fine structure (which is different from the width of each peak) is equal to the reciprocal of the average frequency in the first harmonic, not the overall average frequency in the whole signal. The longer the first-harmonic period, the farther apart are the major peaks in the fine structure. By having harmonic structure in the Signal, the crosscorrelation function's fine structure has a peak spacing that is dissociated from peak width. For example, if the signal's overall average frequency is in the second harmonic, the peak spacing (average first-harmonic period) will be about twice as large as the peak width (second-harmonic period). The artificial increase in peak spacing created by having harmonic structure pushes the side-peaks away from the central peak, and, because they are trapped under the function's envelope, they become less prominent as they move farther away. This trick is exploited by bats that emit harmonically-structured FM echolocation sounds to design crosscorrelation functions having a single prominent peak in the fine structure with no large side-peaks nearby to confuse the process of locating the main peak. These bats use a combination of FM sweeps and harmonics that represents a compromise between a desirably sharp crosscorrelation function and the competing con- straints of atmospheric attenuation of sound and target reflective strength (Hartley 1989; Lawrence and Sim- mons 1982; Pye 1986; Simmons and Stein 1980; Zbinden 1988).

An FM sonar signal has a broad bandwidth that is created by sequential occurrence of frequencies in the FM sweep. This bandwidth, originally distributed across the whole duration of the echo, can be compressed into the much narrower impulse-like crosscorrelation function by the action of the sonar receiver, which, however, must have knowledge of the shape of the FM sweep in the transmitted signal to achieve optimal compression of

the echo. In this sense, the receiver must be matched to the transmitted waveform. Various physical implementa- tions of matched-filter reception of sonar echoes are possible, both in the time domain and the frequency domain, but they all have the equivalent effect of replacing the echo with the crosscorrelation function. After matched filtering of echoes, the receiver then can operate upon the crosscorrelation function to determine echo delay and any other characteristics of the echoes that might be useful. In particular, the receiver must locate the main peak of the crosscorrelation function along the time axis to register echo delay and target range. Reading the crosscorrelation function is a second stage of sonar information processing that relates to the efficiency with which information already represented is actually used (see the a posteriori distribution of Woodward 1964).

Under ideal conditions, the accuracy of locating the peak of the compressed echo - that is, the central peak of the crosscorrelation function - is proportional to the echo's compressed width divided by the detectability index, d= (2E/No) l/z, which is the square root of the signal-to-noise ratio of echoes (Van Trees 1971 ; Wood- ward 1964). For a coherent sonar receiver (one that preserves the phase of echoes and thus the fine structure of the crosscorrelation function), the compressed width of the echo is the width of the central peak of the crosscorrelation function itself, or approximately the reciprocal (period, T) of the signal's average frequency. For a semicoherent receiver (one that destroys echo phase information), the compressed width of the echo is the width of the envelope of the function, or the reciprocal of signal bandwidth (see Menne and Hackbarth 1986). This dis- tinction breaks down in practice for the very wideband FM signals used by many bats. If the bandwidth and the average frequency are comparable in magnitude, the width of the envelope and of the central peak can be about the same (Cahlander 1964).

Target-rangin 9 performance

Assessment of target-ranging performance by bats is conceptually straightforward but technically difficult. The earliest experiments used a two-choice simultaneous discrimination procedure in which 2 identical targets were presented at different distances and bats were trained to choose the nearer of them (Simmons 1973; for a fuller discussion of target ranging see Simmons and Grinnell 1988). In such tests, Eptesicusfuscus could discriminate range differences of 13 to 14 mm (equivalent to 75 to 81 ~ts of echo delay). When the disruptive effect of the bat's head movements during experimental trials was taken into account, Eptesicus could discriminate range changes as small as 5 to 7 mm (29 to 41 ~ts in delay).

The bat's discrimination performance in the early target-ranging experiments could be predicted from the envelope of the crosscorrelation function of the stimulus echoes (Simmons 1973). For 5 different species of bats, the width of the envelope of the crosscorrelation function

- in this case, the reciprocal of the spectrogram bandwidth of the signals - virtually completely accounted for

592 J.A. Simmons et al. : Jittered sonar echoes in echolocation

the observed variance in target-ranging performance (r--0.97; Simmons and Grinnell 1988; Simmons et al. 1975). This prediction amounted to an informal test of whether bats might compress FM echoes into their crosscorrelation functions (Simmons 1969), and the results were interpreted to mean that the image the bat perceives must correspond to the crosscorrelation envelope (Sim- mons 1973). This image relates to the first of the two stages of signal-processing associated with a pulse- compression sonar receiver - the representation of the crosscorrelation function (Woodward 1964). A more thoroughgoing, formal test would include the second stage, too (the a posteriori reading of the crosscorrelation function), by considering echo signal-to-noise ratios as well as crosscorrelation functions (Menne and Hack- barth 1986; Schnitzler and Henson 1980; Simmons 1969; Woodward 1964). However, the more interesting mecha- nistic question concerns whether bats might use a neural network to compute echo crosscorrelation functions from spectrogram-like auditory representations of FM sounds (ARes 1980; 1981; Simmons 1973, 1980). If bats were to achieve nearly ideal target-ranging performance at various echo signal-to-noise ratios, that fact would, of course, be indicative of correlation-equivalent processing, but such a result would say nothing about which correlation-equivalent operation actually is used by the bat. Direct experimental measurements of image shape, particularly relating this shape to such signal representations as crosscorrelation envelopes, provide more thorough characterization of the bat's sonar receiver than is possible just with measurements of accuracy or resolution reduced to a single-valued estimate at one signal-to-noise ratio or another. Both approaches are used in the experiments reported below.

One can assess target-ranging performance independent of echo signal-to-noise ratios by a method that directly measures the image that the bat perceives (Altes 1989). This procedure was developed initially to deal with the artifacts introduced by the bat's head movements during range-discrimination trials, and the results were found to trace the shape of the image of a target (Simmons 1979). In this method, bats are trained to discriminate between electronically-returned echoes that alternate or jitter in arrival-time from echoes that arrive after a stationary delay. During jitter discrimination trials, the bat emits its sonar sounds at rates of 20 to 30 Hz but only moves its head in scanning motions at a rate of about 1 Hz (see Simmons and Vernon 1971). Conse- quently, echoes of two successive sounds, arriving within about 30 to 50 ms of each other, are received before the bat's head has moved very far. When discriminating jittering echoes, Eptesicus can perceive changes in delay smaller than 1 ~ts (Simmons 1979). This submicrosecond value has recently been confirmed in two new jitter experiments conducted by a different laboratory (Menne et al. 1989; Moss and Schnitzler 1989).

The most surprising outcome of the original jitter discrimination experiment was that the bat's performance could be predicted from the fine structure of the crosscorrelation function of echoes rather than just the envelope of this function (Simmons 1979). The implica-

tions of this finding are considerable, particularly for understanding the nature of the acoustic-imaging process that underlies echolocation (Simmons 1980). First, the bat's head movements in ordinary range-discrimination experiments must smear the fine structure of the crosscorrelation function in the data, leaving the results to match only the envelope (Simmons and Stein 1980; Sim- mons and Grinnell 1988). Second, to compute the full crosscorrelation function, including its fine structure, the phase of echoes relative to emissions must be preserved, but physiological data seem to deny this possibility for frequencies higher than several kiloHertz. The bat's actual auditory representation of FM sweeps is in terms of instantaneous frequency (Simmons 1973) or a spectrogram-like pattern of neural discharges (Bodenhamer and Pollak 1981; Simmons and Kick 1984). One can recover the envelope but not the fine structure of the crosscorrelation function by exploiting properties of spectrograms even after phase information nominally is destroyed (ARes 1980, 1981). To recover the fine structure itself, the phase shift between echoes and emissions must be retained. One realistic physiological correlate of phase preservation in such representations might be the time- variability of neural discharges to any one frequency; this variability could be smaller than the period at that frequency (Simmons 1979, 1980). Phase-locking of neural discharges to pure tones at ultrasonic frequencies would not be expected to occur; all that is required is sharp triggering of neural discharges from the envelopes of excitation produced when hair cells are stimulated by FM sweeps moving through their tuning curves. Ob- servations of the time-variance for neural on-discharges to FM sounds in Myotis (Suga 1970), and Tadarida (Bodenhamer and Pollak 1981; Bodenhamer et al. 1979) are as small as several hundred microseconds, and the most recent observations in Eptesieus (Covey and Casse- day 1989) approach the desired condition that this variability be smaller than the period at some frequencies in the bat's sounds (Simmons and Grinnell 1988).

The third implication of the results of the original jitter experiment follows from the reciprocal relationship between the crosscorrelation function and the spectrum of echoes. The formal equivalence of time- and frequency-domain representations might help to explain the bat's ability to combine both echo delay and echo spectral cues into range-axis images (Simmons et al. 1990). Eptesicus can discriminate between echoes whose spectra have been systematically modified by reflection from targets containing surfaces at different distances (Sim- mons et al. 1974; see also Habersetzer and Vogler 1983, for Myotis myotis, and Schmidt 1988, for Megaderma lyra), so target shape (specifically, range profile) most likely is perceived from echo spectra. The results of the original jitter discrimination experiment (Simmons 1979) suggested that bats might create a single range-axis image that incorporates both distance and shape into one perception by expressing all of the information about echoes in terms of the crosscorrelation function (Simmons 1980; Simmons and Stein 1980). Echoes might be modified spectrally by the target's shape, but this modification would show up perceptually in its equivalent time-


d o m a i n fo rm in the c ro s sco r r e l a t i on funct ion . I f the b a t cou ld d i sp l ay neu ra l ly b o t h the spec t rum and the cross- c o r r e l a t i o n func t ion o f echoes ( S i m m o n s et al. 1974), such a t r a n s f o r m a t i o n is p laus ib le . Recen t exper imen t s have now conf i rmed tha t echo spec t ra a re indeed t rans- f o r m e d by Eptesicus in to es t imates o f the s e pa ra t i on o f t a rge t gl ints a long the r ange axis ( S i m m o n s et al. 1990). To be t t e r u n d e r s t a n d this imag ing process , it is h igh ly des i rab le to de t e rmine whe the r the images perce ived by ba t s do in fact c o r r e s p o n d to echo c ro s sco r r e l a t i on func- t ions.

In several respects the resul ts o f the or ig ina l j i t t e r expe r imen t have been conf i rmed in m o r e recent exper i - ments , b u t a full r ep l i ca t ion , wi th a careful e x a m i n a t i o n o f the imp l i ca t ions o f the or ig ina l da ta , has no t ye t been done . Two recent exper imen t s us ing the j i t t e r t echn ique wi th Eptesicus have d e m o n s t r a t e d tha t the fine acu i ty o f less t han a m i c r o s e c o n d is re l iable ( M e n n e et al. 1989; M o s s a n d Schni tz ler 1989). Howeve r , ne i ther o f these two exper imen t s o b t a i n e d d a t a tha t m a t c h the c rosscor - re la t ion func t ion o f echoes. Two other , recent exper i - men t s us ing m o r e s t a n d a r d echo-de lay d i s c r imina t i on m e t h o d s wi th Eptesicus ( S i m m o n s 1973) have o b t a i n e d resul ts cons i s ten t wi th the or ig ina l c ro s sco r r e l a t i on inter- p r e t a t i o n o f the j i t t e r curves (Mas te r s 1989; M a s t e r s a n d J acobs 1989). Howeve r , the n o r m a l d i s c r imina t i on p r o c e d u r e c a n n o t be used to a p p r o a c h the smal l t ime s teps o f the j i t t e r exper imen t s because the b a t ' s head m o v e m e n t s smea r the resul ts (see S i m m o n s a n d Gr inne l l 1988). N o t w i t h s t a n d i n g asser t ions to the c o n t r a r y (Schni tz ler et al. 1985), the or ig ina l d a t a d o in fact con- t a in s ta t i s t ica l ly-s igni f icant peaks in p e r f o r m a n c e tha t a l ign wi th the peaks o f the c ros sco r re l a t ion func t ion o f echoes ( S i m m o n s 1979). (Simi lar ly , the resul ts o f t a rge t - r ang ing exper imen t s wi th different species o f ba t s still s t and as be ing as soc ia t ed wi th the shape o f the enve lope o f the echo c ros sco r r e l a t i on func t ion (see S i m m o n s a n d Gr inne l l 1988), even t h o u g h any po ten t i a l theore t i ca l s ignif icance for this obse rva t i on has been denied (Schni tz ler et al. 1985). W h a t is the difference be tween the or ig ina l j i t t e r expe r imen t and the newer j i t t e r experi- men t s t ha t results in the c ros sco r r e l a t i on r e l a t ionsh ip ' s d i s a p p e a r a n c e f rom the newer d a t a ? Pu t a n o t h e r way, is its presence in the or ig ina l d a t a an ar t i fac t , o r is its absence f rom the newer d a t a an a r t i f ac t ? The w o r k de- sc r ibed be low repl ica tes the cr i t ical deta i l s o f b o t h the o r ig ina l j i t t e r expe r imen t a n d the two newer j i t t e r experi- men t s to answer this ques t ion . The s t imulus cond i t ions used in the ear l ier and in the m o r e recent exper iments differ in a seemingly smal l deta i l t ha t accoun t s for thei r differing resul ts by the c rea t ion o f an u n i n t e n d e d m a s k i n g effect.

Materials and methods

We conducted a series of psychophysical experiments to examine the fine structure of the image of a sonar target as this is perceived by echolocating bats that use FM sonar signals. The animals used in our experiments were big brown bats, Eptesicus fuseus (family Vespertilionidae), obtained from the attics of buildings in Missouri and also in Rhode Island and southeastern Massachusetts. Each bat

was trained in a two-alternative forced-choice procedure to discriminate between an electronically-simulated sonar target whose echoes alternated in delay from one transmission to the next and a simulated target whose echoes had a fixed delay for all transmissions. The bat's task was to detect these alternations in delay, or jitter in echo arrival-time (Simmons 1979; Menne et al. 1989; Moss and Schnitzler 1989), and to respond in the two-choice par- adigm to jittering echoes in preference to stationary echoes. Because echo delay is the acoustic cue for perception of the absolute distance to a target (Simmons 1973 ; Simmons et al. 1990), these experiments effectively present bats with the choice between a target that shifts back and forth in range and a target that is stationary in range. The data reported here consist of the bats' two-choice performance (percentage correct or percentage errors) plotted as a function of the size of the time interval over which the echoes jitter, with additional manipulation of the amplitude, the spectrum, or the phase of one of the jittering echoes relative to the other. These results provide an index of the bat's image of targets along a psychological range axis in perception (Altes 1989).

Target simulation. Figure 1 shows schematically the design of the experiment and the method used to present the bat with electronically-reproduced echoes that simulated sonar targets at different distances by controlling the delay of the replicas of the bat's sonar transmissions that returned to its ears. The bat was trained to sit on an elevated, Y-shaped platform and broadcast its sonar sounds to detect echoes arriving from the left or the right channel of the simulator. One channel was set to deliver echoes that alternated in delay from one value to another (over an interval of At in Fig. 1), thus simulating a target that shifted back and forth from one range to another (a t and a 2 in Fig. 1). The other channel delivered echoes at a constant delay, thus simulating a target at a fixed range (b). This fixed delay was 3.275 ms for return of the echo to the bat's observing position on the platform, which is equivalent to a target range of about 56.5 cm (3.087 ms and 53.2 cm for Bats ~ 1 and ~2). The jittering echoes were delivered around this fixed delay as a mean value. That is, the jitter interval was centered upon 3.275 ms. The bat was rewarded with a piece of a mealworm offered in forceps for every correct choice of the jittering echoes, while no reward was given for incorrect choices of the stationary echoes. The bat's response (arrow in Fig. 1) was to crawl forward along whichever

\ /I \ /

\ simulated t / 1

/ targets /

a 2 t ~ ' ~ b

81 ~ / /

\ / \ / \ / \ /

Fig. 1. Diagram of the two-choice discrimination procedure for studying perception of changes in the delay of echoes that simulate targets electronically. Bats choose between echoes (a 1 and a2) that jitter by a controlled amount (At) and echoes (b) that arrive at a fixed delay. The bat's sonar sounds are picked up at microphones (m), digitally delayed, and then returned to the bat from loudspeakers (s) as echoes. Delay changes are introduced by a controller that resides in the delay system


arm of the Y-shaped platform corresponded to the direction of the jittering echoes. The appearance of the jittering stimulus on the left or right was determined by a pseudorandom schedule and set by a switch located beneath the platform. Control blind trials were run at very small j i t ter values to ascertain that the operator 's knowledge of the switch setting from trial-to-trial had no influence on the bat 's performance. The bat 's ability to determine which channel returns the ji t tering echoes will vary with the size of the jit ter interval (At) in a manner related to the apparent shape of the image of an individual point-target, such as a 1 or a 2 considered by itself (Altes 1989). The experiments were conducted in a 4.5x 3 .5x2 .4 m chamber whose walls, ceiling, and floor were lined with convoluted polyurethane foam (Perma Foam Corp., Irvington, N J) that reduced the amplitude of ultrasonic reverberation by at least 20 to 30 dB compared with what would occur if the chamber had smooth, hard walls.

The electronic system for simulating sonar targets was built around the bat 's observing position on the Y-shaped platform. Two Br/iel and Kjaer Model 4138 condenser microphones (m in Fig. 1) were mounted at the ends of the arms of the platform to pick up the bat 's echolocation sounds. The electrical signal from each microphone was amplified, filtered to the 20 to 100 kHz band of the bat 's signals (see Fig. 17A) with a Rockland Model 442 band-pass filter, delayed by a controlled amount, and then returned to the bat from an RCA electrostatic loudspeaker (Part No. 112343; s in Fig. 1) that was mounted next to the microphone. The distance from the bat to the microphones and loudspeakers was 15 to 20 cm according to the configuration of the apparatus, which was adjusted to minimize extraneous echoes returning from objects near the bat. The path length from the bat to the microphones and back from the loudspeakers was taken into account when setting the delay values for echoes (see below). Also, for bats ~ t and ~ 2, the microphones were located only 12 em away while the loudspeakers were 88 cm away (Simmons 1979). In addition to the bat 's signal being returned as an echo, electronic noise could be introduced at the loudspeaker to manipulate the signal-to-noise ratio of stimuli reaching the bat. Each loudspeaker was supplied with an independent noise source (Elgenco random-noise-generator modules). During representative trials, echolocation sounds recorded from the bats were stored on analog magnetic tape with a Racal Store-4 tape recorder and subse- quently reproduced for digital signal analysis in an IBM PC-AT computer operating with ILS programs from Signal Technology, Inc. Spectrograms of the sounds were made with a Unigon digital sound spectrograph. (On some trials, a QMC Model SM 1 microphone was installed next to one of the simulator 's microphones to record the bat ' s sonar sounds with a better signal-to-noise ratio than is possible using the relatively insensitive Brfiel and Kjaer microphone. At 40 to 50 kHz, this larger microphone is at least 28 dB more sensitive than the B and K model.) Both the simulator microphones and the loudspeakers nominally were located 20 cm (see above) from the bat 's observing position at the center of the platform, so that together they provided a propagation delay of 1.16 ms for any sound emitted by the bat and returning to the bat 's ears after passing through the simulator. The angle separating the two sets of microphones and speakers was approximately 40 ~ .

The bats were tested on an average of 6 days a week, and on the test-days each bat was run on a number of trials that was determined by its current body weight and the quantity of meal- worms consumed after correct trials. Each day's run constituted a block of trials for one of the experimental conditions, which consisted of a particular value of the jit ter interval (At) and often some further electronic manipulat ion of the sounds representing the simulated targets. The bats typically worked through 35 to 60 trials in each block. If the number of trials achieved on a single day was less than this, the same stimulus conditions were repeated the next day to accumulate more trials. A good bat might complete up to 100 trials in one day. The data take the form of percentages of errors or correct responses made over all trials at any particular stimulus condition, and the primary mode of presentation of the data is a graph of percentage errors or percentage correct as a function of the size of the j i t ter interval. For most purposes, the data were not

arbitrarily divided into above and below threshold states because our concern is with the form of the images perceived by the bat as revealed by the shape of the entire curve for each condition (Sim- mons 1973, 1979), rather than with reduction of the data to a single index of discriminability. However, estimates of the smallest detectable jitter are based on a threshold criterion of 75% correct responses (25 % errors) because this quantity has to be stated in terms of some fixed reference. During each trial, the bat moved its head to scan the left and right stimuli, and we monitored these movements for one bat ( # 3 ) using a high-speed (60 frames/s) video system consisting of a Xybion Electronics Systems Model SVC-09 shutter camera (2 ms exposure) and a Gyyr Model 1550 video recorder. The light level in the experiments normally was relatively low, and the bat was visibly disturbed by the brighter lighting required by the video camera, so we attached a Javelin Night Vision Device (image intensifier) to the camera to obtain useable images without added illumination. The practical limit of this system for observing left-right movements of the bat 's head was 0.5 mm.

During individual experimental trials, which lasted for up to several seconds, each bat emitted FM echolocation sounds at rates from as slow as 10 sounds/s to brief bursts of almost 70 sounds/s (see Section on sounds, below). Each sound was received at both simulator microphones (Fig. 1), with an amplitude at each microphone that depended on the aim of the bat 's head during head-scanning movements as the bat searched for the simulated targets (Sim- mons and Vernon 1971). The amplified and filtered signal representing each of the bat 's sonar sounds was delayed electronically by a digital delay system (delay in Fig. 1) designed and built by the Science Services Shops at the University of Oregon (Simmons et al. 1988). The signal was digitized with 12-bit accuracy at a rate of 730 kHz, stored in a circulating buffer memory, and then read out and reconstituted as an analog signal after a preset delay. This analog signal was delivered to one of the loudspeakers for return to the bat as an echo, and it could be mixed with noise to change the signal-to-noise ratio. Additional delay manipulation was done with continuously-variable analog delay lines (Ad-Yu Electronics, Model 801B1) or with pre-cut lengths of R G - 5 8 U coaxial cable. The total gain of the analog circuitry feeding into the delay lines was about 80 dB to bring the peak-to-peak amplitude of the majority of signals to a level just below the 12-bit limit of the digitizer for maximum signal-to-noise ratio. The magnitude of the electronic delay was chosen so that, when added to the 1.16-ms propagation delay from the bat to the microphone and from the loudspeaker to the bat, it created a total delay corresponding to the desired simulated target range. Each microphone-loudspeaker channel was eq- uipped with a dual-delay system, so the bat could be presented with echoes that changed in delay between two values that had b e e n pre-set.

Production of jittering echoes. An electronic controller built into the target simulator created the alternating shifts in echo delay that constituted the stimulus for the bat 's response. This controller served two functions. First, it measured which microphone received the stronger sound for each of the bat 's emissions to determine which side (left or right) the bat was aiming its sound towards. Only the side receiving the stronger sound would return an echo of that particular emission. Second, it selected the delay for any given emission according to the delay that had been presented on the previous emission. For each sonar sound, the bat received only one echo, from the channel towards which the sound was aimed, with a delay that changed back and forth between two values on successive emissions. Figure 2 illustrates the chief components of the jitter system. The electrical signals coming from the left and the right microphones passed through two different systems - a signal path and a control path. That is, they were fed not only into the delay units themselves but also into the jitter controller which determined the value of delay introduced for any particular echo. A remotely-activated switch (left-right switch in Fig. 2) placed the jittering and nonjittering echoes on the left and right sides of the simulator according to the schedule of randomization for experimental trials. Part of the controller's circuitry determined which of


l e f t - r ight sw i t ch

r . . . . . . . . . . . . . . . . . . . . . . . - s ] ~ E - I I de. lav 1 PATH I

left ~ == . ~ . .~= left

IN ~ s e w ~ i r ~ m ixers r l ~ - ! O U T

q

PATH

t h r e s h o l d

Fig. 2. A diagram of the controller that alternates the delay of jittering echoes without introducing extraneous cues for the bat's discrimination. A signalpath introduces electronic delays to signals received by the microphones and switches appropriate delay values to the loudspeakers. A control path counts the bat's sounds to alternate their delays and determines whether the bat aimed its signals at the left or the right channel

the two microphones received the stronger sound for each of the bat's emissions (envelope detectors and comparators in Fig. 2), selected the echo channel to be activated for any given emission (logic gates, 15-ms timers, and flip-flops [ff] following the comparators), and determined the delay value to use for that emission (electronic switches and mixers). The amplitude ratio for the signals at the two microphones was sampled at a point 600 ~ts after the envelope of the bat's sound crossed a pre-set threshold (comparator w/mixer, 600-~ts timer) to determine whether the jittering or nonjittering channel was the one the bat aimed the sound towards. Only that particular channel became active and returned an echo for the emission in question. The flip-flops counted the bat's emissions to the jittering channel (delays 1 and 2) and the nonjittering channel (delays 3 and 4), selecting different delays for successive sounds to create the jittering effect.

The apparatus shown in Fig. 2 changed from one delay state to another and activated the appropriate simulator channel 600 Its after the envelope threshold was crossed. Activation thus typically occurred about 700 to 800 Its into the duration of the sound, which was well before the bat's signal emerged from the delay lines. The timing of activation thus avoided switching artifacts that might interrupt the smooth delivery of each echo and create unwanted cues that the bat could use to perform the discrimination. The 15-ms timers created a window for feeding echoes back to the bat; any sounds occurring outside of this window and picked up by the active microphone thus were not included in the echo waveform. (The bats never emitted echolocation sounds as long as 15 ms.) While one timer was on, the state of the simulator as a whole could not be changed. Each of these timers inactivated the other timer's input to prevent accidental operation of both the left and the right simulator channels should an acoustic transient be picked up and initiate control events while an echo was still being delivered. The delay lines were switched on or off in alternation; their outputs were summed by mixers for the jittering and the nonjittering stimulus even though only one delay line was on at any given time. This insured that the alternating stimuli would pass through common circuitry except for the delay itself. Several different versions of the controller were used with respect to the timing of the change in delay and channel activation relative to the emission itself. In one version, the delay for any given echo was set after the end of the previous emission, and activation of the left or the right channel occurred independent of the delay-selection process. In another, a single 15-ms timer operated both flip-flops rather than a separate timer for each channel to guard against differences between two timers being the source of information for the bat's response. These

different controller configurations all yielded the same data from the bats. The jittering and nonjittering channels were identical except that the alternating delay values were equal for the nonjittering channel and different for the jittering channel. The individual hard- ware delay elements could be interchanged without affecting the data, indicating that no idiosyncracies of one particular delay device were responsible for the bat's performance.

In most of the experiments conducted here, only the digital delay lines were used to generate different delays. These delivered echoes with no detectable delay-dependent amplitude or spectral variations within the limits of accuracy represented by their 12-bit dynamic range. That is, changes in amplitude at any one frequency that might be correlated with delay were smaller than one least- significant-bit of the digitizing process. Separate measurements of the output of the digital delay lines showed no delay-dependent spectral changes larger than 0.03%. This was the practical limit of our ability to observe such changes using differences in transfer functions after the two channels (one storing the input and one storing the output) were balanced in the digital signal-analysis system which performed the transfer-function calculations. In some experiments, to produce delay steps smaller than 1 Its, analog devices for changing delay were placed in series with the digital lines. We used either electronic lumped-constant delay lines that were continuously variable and nominally calibrated to steps of 10 ns, or lengths of coaxial cable cut to produce the desired amounts of delay. From our delay measurements (see below), the RG-58/U cable we used required 35 cm to produce a delay increment of 1 ns.

The actual delay values created by combining different devices were determined directly in two ways. First, steady-state 40-kHz sinusoidal signals were passed through the delay system, and a digital counter operating in a time-interval mode was used to estimate the time separation between the input and the output by averaging many cycles extending over a period of 1 to 2 s. Electronic delays could be measured with an accuracy of about 3 ns by this means. Critical settings of the delay lines were verified using the counter as part of the daily calibration process during experiments (see below). Second, the transfer function of the delay system was measured using 1-ms FM signals (10 to 100 kHz), with a 250-kHz sampling rate for both the input to the delay system and the output. Transfer functions show the delay as a frequency-dependent decreasing phase-shift in the output relative to the input (a negative slope for the frequency-phase plot). Figure 3 shows the change in the phase component of the transfer function of one delay channel set to two alternating delays of 3.275 ms-4-0.5 Its. The jitter interval thus is 1 Its. The reference phase (0 ~ for this graph is the phase for the shorter of the two delays (3.2745 ms). The curve in Fig. 3 shows a linearly increasing phase lag across frequencies that corresponds

0

-0.1 . t ' l~

-0 .2 �84

:K -0 .3

~ - 0 . 4

'~ -o.5

I I I I I I I

i i i i i ~ ^ A A 15 25 35 45 55 65 75 85 95

frequency (kHz)

Fig. 3. A graph of the change in the phase component of the transfer function of the target simulator for an electronic delay change of 1 Its, showing the expected accumulating phase shift (lag) across frequencies for a simple time delay


to the 1-~ts added delay for the longer of the two delays (3.2755 ms). (The graph is read at any one frequency by determining what proportion of the period of that frequency is equivalent to the phase lag and then transforming the phase angle into time.) The phase has to be referred to the shorter delay rather than to the input of the delay system because the total delay of over 3 ms encompasses many 360 ~ phase rotations; a graph of the absolute phase lag would be difficult to read due to the many foldings of the curve as the phase repeatedly wraps around itself in 360 ~ steps.

The transfer function also shows the extent to which the spectrum of echoes varied in a manner correlated with delay. This information is useful to show whether spectral cues might confound the discrimination of jittered delays. Figure 4 illustrates the relative absence of echo amplitude and spectral cues for a series of very small changes in the delay produced by the simulator. For nominal (dial-setting) jitter intervals of 0, 5, 10, 15, and 20 ns, the only feature of the transfer function that changes systematically with delay is the magnitude of the phase lag that increases across frequencies. The amplitude component of the transfer function (Fig. 4A) does not change with delay by amounts larger than about 0.05 dB, and even these minimal changes are not systematically related to the delay value. The phase lag increases with frequency with a slope that is related to the delay value itself (Fig. 4B and C). The actual delays corresponding to these phase lags (and cross- checked with the digital counter method) are 0, 5, 12, 16, and 22 ns. (The amplitude and phase curves all show oscillations that become particularly large for the highest frequencies in the graphs. These are not present in the delay system but instead are artifacts introduced by the process of digitally sampling the input and the output and then computing the transfer function after using a 512-point Fourier Transform with a Hamming window on the data. These oscillations are analogous to the ripple that is present in the frequency response of some types of filters at the edges of the pass-band. Transfer functions are normally displayed with vertical scales of at least 10 to 20 dB for the amplitude component and 360 ~ for the phase component. The ripples appear exaggerated in Fig. 4 because the vertical scales are so much finer than normal.)

"O

r E t~

0 ~ ' ' ' J ' ' '

-0.05

-~ \ . . . . . . 1 i , tO ~ i J

0 Ic t - r

-0.01 15 t ,-

0 - - 0 . 0 2 C n s e c ~ I V ~ I I I i I ~ I I I I I I I i i i i i i i i i i i I i i i i i i i i i i i i

15 25 35 45 55 65 75 85 95 frequency (kHz)

Fig. 4. Graph showing the amplitude (A) and phase (B, C) components of the transfer function of the simulator for changes in delay of 0, 5, 10, 15, and 20 ns. The frequency response of the system is effectively independent of delay, and the accumulating phase lags approximate the expected values (see text). The ripples in the curves are a consequence of digital sampling and analysis; they are not attributable to the delay system itself

Acoustic calibration. The target simulator is an acoustic recording and reproducing system whose performance can be summarized by a frequency-response curve. The frequency response of the left and right channels of the simulator is shown in Fig. 5. These curves refer to the ratio of the acoustic output of the system measured at the bat's observing position relative to the acoustic input delivered to a point 20 cm in front of the bat - that is, at the location of the microphones in Fig. 1. The level-setting attenuators internal to the simulator were placed just prior to the mixers that sum the outputs of the two alternating delay lines (one mixer for the jittering and one for the nonjittering channel - only one delay line is switched on for any given emitted sound; see Fig. 2). These attenuators were set to zero (maximum system gain) for the illustrated frequency-response measurements. At this setting of zero, the absolute acoustic gain of the simulator system at 40 kHz in Fig. 5 was about - 10 dB. In the experiments themselves, the electronic attenuation of echoes was set to produce a lower gain (see below). Calibration was accomplished by replacing the bat with a loudspeaker-microphone assembly. An uncovered Briiel and Kjaer Model 4135 condenser microphone was set at the bat's location, and test signals were put into the simulator by broadcasting 1-ms FM sounds (10 to 100 kHz) from a specially- built electrostatic loudspeaker (Simmons et al. 1979). The system's overall performance was monitored daily by an automatic calibration system built into the target simulator to detect any malfunction of the simulator's acoustic or electronic components. In most conditions, the echo delays for a 1, a 2, and b were set to values in 5-~ts steps with an accuracy of about 1 las. These were verified each day using a digital counter built into the calibration system. Delay values for smaller steps were verified using the counter with sinusoidal signals rather than the usual FM test signals, or by plotting the transfer function determined with FM signals (see phase curves in Figs. 3 and 4). Over a one-year period, the unadjusted gain of the system drifted downward by about 3 dB. The simulator was adjusted daily to maintain its absolute acoustic gain within a 1 to 2 dB span for all experimental trials.

The amplitudes of the echoes delivered to the bat were set at fixed levels with respect to the bat's threshold for detecting a single test target (at the location of b) presented in isolation. The echo- detection threshold was determined for each bat prior to the beginning of the experiments, when the bat was still detecting a single source of test echoes presented either on the left or the right. The amplitude of these test echoes was reduced from about 80 dB SPL peak-to-peak in steps of 5, 2, or 1 dB on successive blocks of trials to find the level at which the bat's performance in the two-choice detection task fell to about 50% correct responses. The two-choice detection threshold was estimated by plotting percent correct re-

.10 L m R .......

" 0 -10

-20

-30 t l

-40

I I I 0 20 40 60 80 100

f requency (kHz)

Fig. 5. Graph of the frequency response of the left and right channels of the target simulator shown in Fig. 1. The sonar signals of Eptesicus contain energy from about 23 to 100 kHz, and the simulator returned the entire first harmonic (60 to 23 kHz) as well as most of the second harmonic up to about 85 kHz


sponses against echo sound pressure level in dB, and the echo level corresponding to 75% correct responses was arbitrarily defined as threshold. In most experimental conditions, the amplitude of echoes representing the targets a 1, a 2 and b was set at 15 dB above this psychophysically-determined detection threshold for each bat. The resulting attenuator settings gave the simulator an overall acoustic gain (from the bat's mouth to the bat's ears) of roughly - 30 dB for the conduct of the jitter experiments described below. The acoustic gain of the simulator in the original jitter experiments was - 34 dB (Simmons 1979). In some conditions (see below) other echo amplitudes (up to 1 or 2 dB greater or less) were used. Finally, the echo amplitudes depended on the strength of the bat's emissions, which consistently varied by several decibels from one bat to another. For all bats, emitted amplitudes were between about 96 and 102 dB SPL (peak-to-peak) at the microphones. These are maxima; the amplitude declined by as much as 6 to 10 dB-during experimental trials as the bat's head aimed from one microphone to the other. During trials with electronic noise added to the echoes to reduce echo signal-to-noise ratios, the two bats tested increased their emissions by about 6 dB.

Specific experimental protocols. Each bat was trained initially to detect and approach a single spherical target presented on the left or the right, then transferred to detecting a single source of electronic echoes (aa) on the left or the right (see Simmons et al. 1990). When this task was learned to at least 90% correct responses, the echoes to be detected were jittered by about 50 to 80 [as while a second set of nonjittering echoes was introduced on the opposite (unrewarded) side. When the bat could discriminate the jittering from the nonjittering echoes at a 50-[as jitter interval, the actual experiments began. The basic jitter discrimination experiment consisted of presenting each bat with a series of different jitter intervals from 50 Its down to zero in steps of 5 [as. (The sonar signals of Eptesicus contain frequencies up to 100 kHz, for which the Nyquist sampling interval is 5 [as. The target simulator is relatively ineffective above about 80 kHz [Fig. 5], so this interval is conservative with respect to the information contained in the echoes.) The basic results were plotted in terms of the percentage of correct responses as a function of the size of the jitter interval. Next, thefine acuity of jitter discrimination was assessed by making jitter intervals of decreasing size from 5 [as down to zero in steps as small as 5 ns. The fine-acuity results also were plotted as percentage correct responses for different jitter intervals, to determine the 75-percent-correct threshold for jitter discrimination. (Step sizes smaller than 5 [as were needed in case the bat could make delay estimates of fractions of the Nyquist sampling interval.)

Several more stimulus manipulations were then incorporated into the experiments to refine the data with respect to identifying the auditory basis for jitter discrimination. The first refinement of the experiment examined the ability of the bat to perceive information conveyed by the phase of the echoes relative to the emissions. The original jitter experiment suggested that bats might perceive echo phase as a part of the process of image formation (Simmons 1979; Simmons and Grinnell 1988), and the specific phase-shift experiment used here has been proposed as a means of resolving the question (ARes 1981; Menne and Hackbarth 1986). Recent experiments have demonstrated that Eptesicus can discriminate echoes jittering only in phase (+ 45 ~ and - 4 5 * or 0 ~ and 180~ Menne et al. 1989), which is to be expected from the original result that the fine structure of echo crosscorrelation functions governs performance. However, a crucial aspect of the phase question is whether shifts in echo phase can be made to trade with corresponding shifts in echo delay. One experiment using ordinary delay discrimination rather than jittered echoes has shown phase-time trading to occur (Masters 1989). The recent study using jittered echoes failed to sample small enough time-shift steps in the region of the expected location of the main peak in the data to determine whether time and phase trade off for a + 45 ~ phase-shift, and only a single time value was used for a 180 ~ phase shift (Menne et al. 1989). (Specifically, a • 45 ~ phase-shift should displace the low-performance peak in the data from zero to about 7 ~ts, while this peak itself is less than 0.8 ~ts

wide [Menne et al. 1989]. In the recent experiment, jitter steps of 4 and 8 Its were used, but nothing was sampled close enough to 7 Its for the peak to appear in the data if it were present. Time-phase trading cannot be tested without sampling an appropriate range of delays, and, for the 180 ~ phase shift, only one value was actually tested. Thus, although the time/phase question was posed, this recent experiment never examined conditions which would answer it.) In our experiment reported here, the phase of echo a t was inverted by 180 ~ relative to echo a 2 to determine whether the bat might perceive the phase of echo waveforms as part of the structure of images. This phase shift was accomplished by inserting a gain of

- 1.0 into the delay path for at, so that alternation of delay by the amount At was accompanied by alternation in phase. The size of the jitter interval was changed in 5-[as steps; the experiment simply assessed the impact of the phase shift on perception of the jitter for different amounts of jitter. The 180 ~ phase shift produced no artifactual amplitude or spectral changes that could be detected within the 12-bit accuracy of the transfer-function calibration procedure described above.

In the second experimental refinement, the bandwidth of the echoes returning to the bat was restricted to the region of 55 kHz by setting the upper and lower band edges to these values on the Rockland Band-Pass filters that preceded the simulator delays (Fig. 2). In the normal condition, the frequency response of the

' simulator system is as shown in Fig. 5, but in the 55-kHz filtered condition, the frequency response peaked at 55 kHz, with 24 dB/ octave roll-offs above and below this value. The 3-dB points on the upper and lower slopes of this restricted band are at 50 and 60 kHz. The basic jitter experiment was conducted in the 55-kHz condition, as was the 180 ~ phase-shift experiment (see above).

In the third experimental refinement, the amplitude of echo a 1 or echo a 2 was changed slightly (typically by 1 dB) to determine whether the apparent delay of that echo shifts by an amount predicted from the amplitude-latency trading relation for N1 and N 4 auditory neural responses in Eptesicus ( - 13 to - 18 [as/dB; Simmons et al. 1990). This procedure can be used to distinguish between acoustic cues encoded by the timing of neural discharges at each frequency and cues encoded only by the specific frequency to which neural channels are tuned - that is, as spectral information. If the bat encodes the arrival-time of individual echoes in terms of the time-of-occurrence of neural discharges in the ascending auditory pathway (Bodenhamer and Pollak 1981; Simmons and Kick 1984; Suga 1988), the observed physiological latency shift should translate into a displacement of the bat's basic jitter discrimination performance curve along its horizontal time axis by approximately 13 to 18 Its/dB, in a direction dependent on whether echo a~ or echo a 2 is changed in amplitude. The occurrence of such a shift in the jitter behavioral data would demonstrate time-intensity trading, indicating that the bat uses a temporal representation of echo delay rather than an artifact-related spectral representation to encode the jitter. The absence of such a shift would implicate spectral cues or other non-time cues as the basis for jitter discrimination.

In the last experimental refinement, broadband electronic noise was added to the echoes delivered to the bat to assess the effect of reducing the signal-to-noise ratio of echoes upon the bat's ability to perceive very small changes in echo delay. An independently- generated noise was delivered to each of the loudspeakers to avoid complications that would be introduced by having correlated noise in the left and right channels. This added noise exceeded the noise level initially set within the system by the sensitivity of the two microphones. Having measured the acuity of jitter discrimination in quiet conditions (see above), the procedure now was to introduce broadband noise to the stimuli, beginning at a low level and increasing the level in steps of about 2 dB, until the bat's jitter acuity began to decrease as a consequence of the added noise. To insure that the data for estimating the limiting effects of noise on the bat represented a condition in which the bat's performance was in fact limited by the external noise, the noise level was increased until the bat's jitter acuity was made 3 to 4 times worse than in the quiet. The signal-to-noise ratio at this reduced acuity was then determined from recordings of the echo and noise signals.


The echo signal-to-noise ratio was measured by digitizing the signal and the noise separately (500 kHz sampling rate; 12-bit accuracy) at the input to the loudspeaker and determining the ratio of the total integrated energy under the envelope of the signal alone to the RMS noise energy per Hertz, or mean-squared noise amplitude per Hertz. The result was then expressed as the detectability index, d = (2E/No)l/2 or the square-root of the echo signal-to-noise ratio (Menne and Hackbarth 1986; Woodward 1964) for representative sonar sounds from each bat. These calculations on digitized signal and noise waveforms were performed using ILS signal- processing routines (Signal Technology, Inc.) in an IBM PC-AT computer. Each digitized sonar signal was full-wave rectified and low-pass filtered (5 kHz) to form its envelope, which was then integrated over signal duration to determine relative signal energy. The noise was full-wave rectified to determine relative mean-square amplitude or RMS energy, which was then apportioned over the 80-kHz signal or noise band to determine RMS noise energy per Hertz. In the experiment, both the signal and the noise (which was flat to within + 1 dB) initially were filtered to the band from 20 to 100 kHz prior to being broadcast from the loudspeakers, which then delivered signal and noise spectra corresponding to the frequency responses shown in Fig. 5. The noise was continuously delivered at the scheduled amplitude throughout the block of trials run at any particular noise level.

To evaluate the bat's performance relative to a coherent ideal receiver, the expected acuity for an ideal receiver was first calculated from Woodward's Equation (Woodward 1964; Menne and Hack- barth 1986),

a t = (T/2rtd)

where a t is the expected standard deviation of the delay estimate, T is the period of the central peak of the crosscorrelation function of echoes (18 las; approximately equivalent to the reciprocal of the average frequency of echoes or the RMS bandwidth, which is about 55 kHz; see Fig. 17A in relation to Fig. 5), and dis the detectability index or the square-root of the echo signal-to-noise ratio as described above. Going beyond this standard equation, the expected acuity was then expressed in terms of predicted hyperacuity, kvs= 1.92 at, as specifically derived for jitter-discrimination data (Altes 1989). Expected jitter acuity also was separately determined using a graph derived from computer simulation of jitter-discrimination trials (Menne and Hackbarth 1986, Fig. 12). The intention was to determine whether the bat's noise-limited acuity conformed to the performance expected ofa crosscorrelation receiver operating on the bat's signals at the signal-to-noise ratio prevailing in this particular experiment. For comparative purposes, expected hyperacuity was also calculated for a semicoherent receiver under the same conditions, using the value of 13.9 kHz for bandwidth (centralized RMS bandwidth) employed in the computer simulations of jitter trials (Menne and Hackbarth 1986).

Results

Basic jitter-discrimination performance

A total o f 6 Eptesicus have now successfully completed the basic ji t ter discr iminat ion experiment for jitter intervals f rom 50 ~ts d o w n to zero in steps o f 5 ~ts (some time steps were 10 Ixs for one bat, shown as Bat 44=2, below). Two o f the bats (Bat ~ 1 and Bat =14= 2) were run in the original j i t ter exper iment (see procedures in S immons 1979), and 4 bats (Bat ~ 3 th rough Bat ~ 6 ) were run under the condi t ions described here. Figure 6 shows the percentage o f correct responses achieved by all 6 o f these bats, with 40 to 60 trials per data-point .

The curves for all 6 bats have the same shape, as described in the original j i t ter discr iminat ion experiment

lOO

~ g0 0

~, 80 n-"

7o

0 o 60 *d

~ 5o

I1; . . . . Bot . / j / m - - m B a t #.3

* - - * Bat #4 ~/ *-------* Bat #5 g

o - - o Bat #6

t t i i I i i i i r I

0 5 10 15 20 25 30 35 40 45 50 Time (/zsec)

Fig. 6. Performance (percentage correct) of 6 Eptesicus in the basic jitter discrimination task. Bats ~ 1 and ~ 2 are from the original experiment with a different placement of the loudspeakers (Sim- mons 1979), and Bats ~ 3 - ~ 6 are from the present experiment. Each data-point is 40-60 trials

(Simmons 1979). The per formance o f the bats is most ly in the range o f 90 to 95 percent correct responses for jitter values o f 5 ps or more, with a consistent region of somewhat poorer per formance at 30 to 35 Ixs. Each o f the 6 bats unambiguous ly exhibits this poor -pe r fo rmance notch. Pool ing data f rom 5 o f the 6 bats (Bat =14= 2 missed the 35-~ts point), the binomial-dis t r ibut ion probabi l i ty that per formance at 35 ps would be as low as 85 % correct by chance (relative to per formance at 10 to 25 ~ts and at 40 to 50 ~ts) is less than 0.001. Thus, the local d rop in per formance is a real feature o f the data. N o n e o f the bats performs differently f rom chance (50 % correct) at a jitter value o f zero, indicating that the bats are no t discr iminat ing some incidental characterist ic o f the experimental setting but instead perceive some stimulus feature explicitly related to the setting o f the delay lines.

Fine acuity for jitter discrimination

The per formance o f the bats declines abrupt ly in Fig. 6 f rom most ly 90 to 95% correct responses at 5 gs to a round 50% correct at zero. The threshold limit o f jitter discrimination by Eptesicus must fall somewhere in this interval. Previous experiments have variously est imated this limit to be less than 0.6 gs (Simmons 1979 for Bat

1 and Bat 4~2) and less than 0.4 gs (Menne et al. 1989; Moss and Schnitzler 1989). Three o f the bats used in our jitter experiments (Bat 4~ 1, 4~ 3, and 4~ 5) were tested for jitter discrimination at smaller intervals than were possible in previous experiments, and all 3 yielded estimates o f a closely-measured minimum-detect - able jitter value. Figure 7 illustrates the per formance o f these 3 bats discriminating jitter intervals f rom 60 ns, or 0.06 ~ts (1 n s = 10 - 9 s), down to zero in steps o f roughly 5 ns. These small delay changes were p roduced either with delay lines or by varying the length o f coaxial cables in the simulator. The curves in Fig. 7 are closely aligned regardless o f the individual bat or the me thod used to change echo delay. Using 75% correct responses as the criterion for discrimination, Eptesicus can perceive

J.A. Simmons et al. : Jittered sonar echoes in echolocation

100

o 80

. / . , ~ Delay line n- 70 ~1 V - Bat # 1 ,-- �9

~ , , Bat # 3 w-- -m

6o o / / / Cable

a ] Bat # 3 o - - - a o ~ 50 ~ Bat # 5 o

Y 40 ' ' ' ' ' ' ' ' ' ' , , ,

0 5 l O 15 20 25 30 35 40 45 50 55 60

Time (nanoseconds)

Fig. 7. Performance of 3 bats discriminating echoes that jitter by very small amounts. Average acuity of the bats for jitter is about 10 ns. Two different delay-generating techniques were used - electronic delay lines and cables of calibrated lengths. Each data-point is 40-60 trials

changes as small as 10 ns in the conditions of this experiment. The procedures for calibrating the delay system (see Fig. 4) seem to rule out any equipment-introduced spectral artifact as a cue for perception of such small time shifts. The bats evidently were discriminating some stimulus feature explicitly related to the delay of the sounds passing through the simulator. Possible echo cues will be examined below.

Effects of echo phase-shift on jitter discrimination

The presence of a 180 ~ phase shift on echo al dramati- cally altered the form of the jitter discrimination curve obtained for each of the 4 bats tested on both 0 ~ and 180 ~ phases of al. The curves shown above in Fig. 6 represent jitter-discrimination performance in the 0 ~ phase condition for all 6 bats. The same 0~ data are shown as percentage errors in Fig. 8 for the four bats now tested on both phase conditions (0 ~ data in upper part of graph). The performance obtained from these same 4

599

bats in the 180~ condition is plotted upside-down in Fig. 8 (180 ~ data in lower part of graph) to show the change in the curves that is associated with the phase shift. (The vertical percentage-error axis is simply reflected around its zero origin.) The primary peak in the error curve at zero and the secondary peak at 30 to 35 ps for the 0 ~ phase condition are transformed into a peak at 15 ~ts and a suggestion of a peak at 45 Ms for the 180 ~ phase condition. Pooling the 180 ~ data for all 4 bats, the binomial-distribution probability that the 15-~ts peak in the error curve would deviate by chance from the performance prevailing at most other time values is less than 0.001. The probability that the 45-~ts peak would occur by chance is less than 0.05. The difference in the appearance of the curves for 0 ~ and 180 ~ phases of echo al relative to a 2 thus reflects a real difference in the performance of the bats in these two conditions.

A modification of the graph showing jitter-discrimination performance in the two phase conditions high- lights the nature of the phase-shift effect. The dual plot in Fig. 8, with performance for 00 phase on top and inverted performance for 180 ~ phase on the bottom, can be converted into a compound curve that combines the two sets of data in a manner similar to that used for compound period histograms of auditory-nerve responses (Kiang et al. 1965; Goblick and Pfeiffer 1969). Figure 9 shows the mean performance for the 3 bats (whose 0 ~ side-peak is at 35 ~ts; Bat :~ 3, ~ 4 , and ~ 5) on 0 ~ and 180 ~ phases (open circles), with the extreme values above or below the zero origin of the percentage-error axis connected by a common or compound curve (solid circles). I f the percentage-error scores above and below this zero origin at any given time value do not differ by more than 2% errors, the compound curve shows its data-point at zero. Otherwise, the compound curve traces the larger of the upward or downward deviations away from the zero origin of the vertical axis. Figure 10 shows the individual compound curves for each of the 4 bats that completed the phase-shift experiment. These compound curves show that the phase-shift experiment had a similar effect on each bat. Because the calibration procedure

60 Bat #3 m--w

50 ~ Bat #4 "- -" O~ Bat #5 � 9

40 ~ Bat #6 ----

\\ 3O

~ 20 i , i

-~ 10

o ~ o -

a_ 10

20 \,j 180~ 30 ' ' ' ' ' ' ' ' ' ' '

0 5 I0 15 20 25 30 35 40 45 50

Time (/zsec)

Fig. 8. Performance of 4 bats (percentage errors) discriminating jittering echoes with a phase-shift of 0 ~ or 180 ~ between echoes. Vertical scales for the two phase conditions are aligned with a common zero origin to better illustrate the nature of the phase effect. Each data-point is 40-60 trials

50 �9 O~ o - - o o

40 \ 180~ o - - o 0 phase

30 \ compound � 9149

B 2O

u3 lo

o - y_---oz -_%T - -

20 180~

30 ' ' ' . . . . . ' ' , 0 5 10 15 20 25 30 35 40 45 50

Time (microseconds)

Fig. 9. Illustration of the construction of a compound performance curve from 0 ~ and 180 ~ phase-shift data. The rationale for using a compound behavioral curve is the same as for a compound period histogram in auditory physiology (Kiang et al. 1965; Goblick and Pfeiffer 1969). Values are means for 3 bats (see text)

600

6o

2 ,.h

Bat #3 ~ - - i OOphase 50 ~ Bat #4 � 9 1 4 9

A Bat #5 * - - * 40 ~ \ Bat #6 o - - o

\\ 3O

0 . . . . . . . . . . . . . . . . . .

10 ~ - ~ - - ~ / ~ ~ ~

20 \ / "~ 180~

30 ' ' ' ' ' ' ' ' ' ' ' 0 5 10 15 20 25 30 35 40 45 50

Time (#sec)

Fig. 10. Individual compound jitter discrimination curves for 4 bats. The cyclic shape of curves is independently present in the data from all 4 bats

i - - m Bat #3 u'~~

,304050 ~ l t e r

2 20

~ o

20 ~ 180 phase

3 0 ' ' ' ' ' ' ~ ' ' ' '

0 5 10 15 20 25 30 35 40 45 50

Time (microseconds)

Fig. 11. Performance of a representative bat on jitter discrimination with echoes filtered to a 55-kHz band, with phase-shifts of 0 ~ or 180 ~ between jittering echoes. Each data-point is 60 trials

uncovered no evidence of a spectral artifact introduced with the phase shift, the effects of shifting the phase of echo ax relative to a2 depend on a specific stimulus feature associated with different delays of the sounds passing through the simulator. The further significance of the compound performance curves will be discussed below.

Effects of 55-kHz filtering on jitter discrimination

Bats transferred from the broadband basic jitter task to discrimination of jittering echoes filtered to a narrow band around 55 kHz with no difficulty. Figure 11 shows the performance of one bat (Bat # 3 ) discriminating 55-kHz filtered echoes with either a 0 ~ or 180 ~ phase-shift added to echo a~. The locations of the error peaks in the 0 ~ and 180" performance curves are moved to different time values along the horizontal axis compared to broadband jitter performance (see Fig. 8). The compound performance curve for the same bat in the 55-kHz filtered condition is shown in Fig. 12. The shifted locations of the error peaks in the filtered relative to the unfiltered conditions are evident through a comparison of Fig. 10 with Fig. 12.

s

J.A. Simmons et al. : Jittered sonar echoes in echolocation

50 i - - i Bet #3 T O~ 55 kHz filter

0 - - - -

10

2o ,30 180~

40 ' ' ' ' ' ' ' ' ' ' ' - 5 0 -4-0 - 3 0 - 2 0 - I 0 0 I0 20 30 40 50

Time (microseconds)

Fig. 12. Compound jitter discrimination curve for echoes filtered to a 55-kHz band (see Fig. 21)

0 d B o - - - - o

40 +1 dB a2o---o

e 30 ~ " I t I1) ~ u ;

20 i , ===:> ,x ,

~ca 10 : .6 ',, ! , , b., �9

I . ~ ~ ~ ~. ..... . - ; ~ / -~ , I . . . . . . . . . . . . . . .

O h , V " ' , , , , 0 10 20 30 40 50 60

time (}Jsec)

Fig. 13. Graph illustrating amplitude-latency trading for jitter discrimination performance of Bat 4~ 3. Curves shift by about - 17 las/dB whether the earlier of the jittering echoes (al) is decreased by 1 dB or the later of the echoes (a2) is increased by 1 dB. The shift is predictable from neural amplitude-latency trading data (see text). Each data-point is 60 trials

50 Bet #30dB =- - o

4-0 t~ - l d B m- -m O~ hose

30

10 ~ ~m

~ o . . . . . . . . . .

o_ 10

%/ ,80Op.ose 30 ' ' ' ' ' ' ' ' ' ' ' ' '

0 5 10 15 20 25 30 35 40 45 50 55 60

Time (microseconds)

Fig. 14. Amplitude-latency trading effect for the compound jitter discrimination curve of Bat ~ 3. The entire cyclic shape of the curve is displaced to the right by about 17 gs when echo a 1 is decreased by 1 dB. Each data-point is 60 trials

Effects of echo amplitude shifts on jitter discrimination

Two bats completed a series of jitter discrimination experiments in which the amplitude of either echo al or a2 was changed by 1 dB with respect to the amplitude of the other echo. Figure 13 illustrates the effects of decreasing the amplitude of a, or increasing the amplitude of a2 by


5 0 Bot # 5 0 d B + - -<+

40 '~ - l d B * - - * O~ I

30

20 ~ , \ I i o / \ @ ~ \ o , . / \ ; . ~ . %/ o\ ,,~ 10 " / \

.: o _ _ - , . . . . L . . . . . ~ / / \ / i * / " - /

20 180~

30 . . . . . . . . . . . . . 0 5 10 15 20 25 30 35 4-0 4-5 50 55 60

Time (microseconds)

Fig. 15. Amplitude-latency trading effect for the compound jitter discrimination curve of Bat # 5. Again, the curve is shifted to the right by 17 las. Each data-point is 40-60 trials

this amount. In either case, the basic (0 ~ phase) jitter discrimination curve shifts to the right (to later time values) by about 17 laS while otherwise preserving its original shape. The compound performance curve derived from both 0 ~ and 180 ~ phase-shift data (see Fig. 9) is similarly displaced along the time axis by a change in echo amplitude. A 1-dB decrease in the amplitude of at moves the entire compound performance curve to the right by roughly 17 p.s, equally affecting both 0 ~ and 180 ~ components. Figures 14 and 15 show this rightward shift in the compound curve for the two bats tested on both phase and amplitude changes of a~.

Aside from being displaced in time, the only difference in the appearance of the curves before and after the 1-dB echo amplitude change in Figs. 13, 14, and 15 is a reduction in the height of the main error peak that shifts from zero to about 17 ~ts on the time axis. This reduction is a consequence o f the extreme narrowness of the main peak - it is only about 20 ns wide (Fig. 7), so the exact tip of the peak can only be found after the amplitude change has been imposed by sampling a sequence o f jitter values along the time axis in steps of about 5 ns or less. At a normal rate of 40 to 60 trials per day, the search process would require about a year of experiments just to cover a span of 2 Ixs (2000 ns), which would be wasteful of the valuable resource constituted by the trained bats. Hence, the main error peak has been located only approximately here by using a small number of trials to home in on its general location, where a full 40 to 60 trials were then conducted. The 0~ side peak at 30 to 35 ~ts and the 180*-phase side peak at 15 l~S are each very broad and do not suffer partial disappearance after the echo amplitude change when the time steps are as large as 5 Ixs. Consequently, they appear about the same after being moved sideways. (The locations of these side peaks in the shifted performance curve help to estimate the region to be searched for the much narrower main error peak that originally was located at zero on the time axis.)

Echolocation signals for jitter discrimination

During individual trials of the jitter experiments, each bat emitted a series of FM echolocation sounds that

emission -- 1'2t 34 56 r,a 91017!2r~4i,711~0~24 2 ! 2r. 217 2is ~ 3 2

. I G n r ' , ,l . . . . . . . r. l l , l, Ul l lq ' I I I l i i, , , , �9 , , 1 11 I t c

echo r l ' , , : ~ ,;~ ~oU24' r I ~r~ 1 2 3455 28

26 7 1̂ 13 15 18 22 24 25 I ~ 28293031

. . . . . . . . . i th it i l , ~I. J . . . . . emission �9 1 2 +34_0.6 8 910 11 z I " r . . . . . . . "1 r | ' l l i , l ~ ' i r i I " [ I ' " ' +

L E F T e c h o . I , a I . I.

13 -15 27

I m o v e s f ~ 1 s c a n s

I m a k e s c h o i c e f R )

I 1 sec I

Fig. 16. Sonar emissions recorded from Bat ~ 3 during a discrimination trial with 15 ns jitter. The 4 traces show the sonar emissions at the right microphone, the echoes returned on the right, the emissions at the left microphone, and the echoes returned on the left. The bat emitted 32 signals in all, first while it climbed onto the platform towards the right (sounds 1-10) and then while it scanned to the left (sounds 11-18), the right (sound 20), the left (sound 22), and the right (sounds 23-24). The bat then moved to the right arm of the platform after making its choice, receiving several echoes from one side or the other (sounds 25-32) as it scanned for the mealworm offered in forceps

covered a frequency range from about 23 to 100 kHz, as is typical of Eptesicus fuscus during the approach or tracking stage of pursuit (see Simmons 1989). The sounds were emitted in a pattern that revealed a structure to each discrimination trial. Figure 16 is a recording of all the sonar sounds emitted and echoes returned during one trial with Bat # 3. The 4 traces in Fig. 16 show the sounds picked up at the microphones (left and right emissions) and the echoes delivered back to the bat by one or the other channel of the simulator (left and right echoes). The bat received 8 echoes from the right channel while it moved forward onto the platform, 3 echoes from the right and 8 echoes from the left during the scanning stage, and a few additional echoes from both sides after it had made its choice and moved to the right side of the platform, which was a correct response. The scanning stage was distinguished by particular attentiveness on the part of the bat to the stimuli, with smaller movements of the body and head and deliberate aim of the head towards the microphone and loudspeaker on one side or the other. By keeping the simulator switched off until the bat had climbed onto the platform for Bats # 1 and # 3, we could exclude the use of echoes arriving outside of the middle, scanning stage of the trial. Both coarse and fine acuity were unaffected by this precaution, indicating that the scanning port ion of the trial, with reception of no more than 20 echoes overall, was sufficient to yield the performance shown in Figs. 6 and 7.

In practically all trials, the interval between sounds started at about 50 to 100 ms (repetition-rates of 10 to 20 sounds/s) as the bat moved forward onto the platform, decreased to as little as 15 to 30 ms (repetition-rates of 30 to 70 sounds/s) during the middle port ion of the trial when the bat scanned to the left and the right to make its choice, and then increased again to about 50 to

602 J.A. S immons et al. : J i t tered sonar echoes in echoloca t ion

100 ms as the bat moved onto one or the other arm of the Y-shaped platform after having decided which echoes were jittering. During the scanning stage, shorter intervals of 15 to 20 ms or so were common for trials where the magnitude of the jitter was less than 1 gs. These sounds were emitted as several bursts of 3 or 4 signals per burst (see Fig. 16) over a span of about 200 to 500 ms, and the bat typically held its head more stationary during this time than at other times in the trial. We used high- speed video equipment to observe the position of the head of Bat ~ 3 during 20 trials in which the jitter was only 20 ns and found that the bat moved its head laterally by as little as 0.5 mm and by as much as 1 to 3 mm during the interval of 15 to 30 ms between successive sounds. These lateral head movements translated into movements between successive sounds of about 0.3 to 1 mm towards or away from the microphones and loudspeakers. In contrast, the delay acuity of 10 ns corresponds to a displacement of only about 0.002 mm in range. At relatively large jitters (5 gs and more), the bats appeared more casual about trials, with larger movements of the head while scanning, and longer, more irregular intervals between sounds. Bat ~ 5, which used longer sounds than any of the other bats, also emitted its sounds with somewhat longer intervals between sounds than the others, but it still exhibited a shortening of the interval to as little as 30 ms when it scanned the simulated targets for very small magnitudes of jitter.

The total number of sounds emitted during a trial depended primarily on whether the bat made its choice after only one scan of each side or after several scans. Typically, the total was about 8 to 20 sounds actually passed through the simulator to become echoes, with about 2 to 5 echoes received for each of the jitter values presented. That is, delays al and a2 would each occur 2 to 5 times before the bat made its choice, and delay b would occur up to 10 times. It should be emphasized that there were many trials where only two echoes were recorded for each delay value, and yet the bat responded correctly. In some instances, only one echo per delay was delivered. Figure 17A shows spectrograms of representative sounds recorded from Bats ~ 3-~ 6, and Fig. 17B shows distributions of durations for all 6 bats on several trials. The sonar sounds of Bats ~ 3 and ~ 4 were quite similar, being 1 ms to slightly under 2 ms in duration, with a prominent first and second harmonic and a somewhat less prominent third harmonic. Bats ~ 1 and ~ 6 used similar signals but with durations up to 2.3 to 2.5 ms. In contrast, Bat 4~ 2 emitted sounds as long as 3.3 to 3.5 ms, and Bat ~ 5 emitted sounds that were from 3 to 4 ms in duration interspersed with shorter sounds with a duration of about 2 ms. The third harmonic was absent in these longer sounds and there was no overlap of the first and second harmonics at 55 to 60 kHz. These changes are typical for sounds longer than 2 to 3 ms in Eptesicus and are thus not peculiar to bat ~ 5 but are just a consequence of the sounds being longer. Nevertheless, Bats ~ 2 and ~ 5 performed in the same way as the other bats in all o f the experiments. Particularly significant is the presence of identical amplitude-latency trading in the data for both Bat ~ 3 and Bat ~ 5 in Figs. 13-15. The

A

100

80

60

40

3 5 6 4

J ~ J J 2 4 5 time (ms)

B

1 3

l l l l l l l l ! 4 o 1 2 3 4 ~ ~ _ - - ~

Fig. 17. (A) Spectrograms of representative sonar sounds recorded from Bats # 3-:~ 6 during data-collection trials of the basic jitter discrimination experiment. Time scale refers to the individual spectrograms; spacing between them is not related to actual repetition-rates used by the bats. (B) Envelopes of sonar signals emitted by Bats ~ 1-6 during jitter discrimination trials

increased duration of most of the sounds emitted by Bat 5 did not alter the dependence of the jitter data on the

latency of neural discharges compared to a bat that used shorter sounds for jitter discrimination.

Effects of echo signal-to-noise ratio on jitter discrimination

Two bats that had completed the fine acuity measurements in quiet conditions (Bat ~ 3 and # 5 in Fig. 7) were induced to continue discriminating small jitter intervals in the presence of increasing levels of broadband masking noise delivered continuously from the two loudspeakers. The performance of the bats began to deteriorate from an acuity in the quiet of 10 ns when the RMS amplitude of the noise approached about one-fifth of the peak-to- peak amplitude of the echoes. At this point, the bats lengthened their sonar emissions to about 4 to 6 ms from the shorter durations they used in quiet conditions (see above). Further increases in the noise level reduced each bat's performance until fine acuity was about 4 times worse than in the quiet, thus insuring that the noise itself was the limiting factor to the bats. In this noise condition


B IE d . . . . B

I I

10 msec

Fig. 18. Waveforms of 4 representative echoes delivered to Bat # 3 during the jitter experiment with added noise. Each waveform shows the echo emerging from its noise background. Values of detectability index (d; see text) given in dB

and for a jitter discrimination of 40 ns, Bat # 3 achieved 77% correct responses (60 trials) and Bat # 5 achieved 67% correct responses (60 trials).

Figure 18 shows 4 representative echoes in broadband noise recorded from Bat # 3 during jitter-discrimination trials with a jitter interval of 40 ns. The computed detectability index (d), or the square-root of the signal-to-noise ratio (see Methods), is given beside each echo waveform. The detectability index for 50 sample echoes delivered to the bats ranged from about 56 (35 dB) to a maximum of 68 (37 dB) for the strongest signal sampled. As a check on these values, a rough approxima- tion to the value of d can be made directly from the waveforms in Fig. 18. By eye, the RMS noise amplitude in these examples is about one-fifth of the peak-to-peak amplitude of the echoes (2A), which itself is twice the zero-to-peak amplitude (A). RMS noise amplitude thus is roughly 0.4A. The relative echo energy, E, can be estimated from A2• duration, using a value of 5 ms (5 • 10-3 s) as the approximate duration of a rectangular envelope with the same mean energy as the actual echoes returned to the bat. Thus, E=5A2 x 10 -3. The overall relative noise energy is approximately the square of the RMS noise amplitude (0.4A) 2, which must be distributed over a noise bandwidth of 80 kHz (20 to 100 kHz) to obtain the approximate relative value for No of 2A 2 • 10-6 per Hertz. The resulting estimate of the echo signal-to-noise ratio, 2E/N0, is thus (10A2x 10-3)/ (2A2• 10-6)=5000. The detectability index, d, is the square-root of the signal-to-noise ratio, which is 70.7, or about 37 dB.

The values obtained for 29 of the 50 signals were between 60 and 63 (36 dB), so the waveforms shown in Fig. 18 are in fact both typical of the detectability indices that were available to the bats during this experiment and indicative of the largest detectability index that would be available under the experiment's conditions. The gain of the simulator had previously been adjusted so that the amplitude of those echoes that were just barely clipped to the maximum allowed by the simulator would correspond to the amplitude shown for the echo with a detectability index of 37 dB in Fig. 18. Thus, sonar emissions of greater amplitude would not yield echoes of corre- spondingly greater amplitude. The durations of signals emitted in the noise experiment by the bats were never observed to be longer than 6 ms, so there was no practical

Table 1. Predictions of jitter discrimination for coherent and semicoherent echo processing at different values of the detectability index, d. (RMS bandwidth=55.3 kHz; centralized RMS bandwidth= 13.9 kHz)

d Coherent Semicoherent

3.16 (10 dB) 1.75 laS 3 Its 6.96 las 7 las 10.0 (20 dB) 555 ns 500 ns 2.20 las 1 ~ts 31.6 (30 dB) 170 ns 100 ns 700 ns 700 ns 63.2 (36 dB) 87 ns 50 ns 348 ns 400 ns

a Altes (1989) b Menne and Hackbarth (1986)

opportunity for the bats to receive echoes with detectability indices higher than about 37 dB.

We take the most typical value of 36 dB to be the echo detectability index associated with a fine jitter acuity of 40 ns. The Woodward Equation (above) predicts that this value should yield a standard deviation for delay estimates of about 45 ns for a coherent receiver operating on crosscorrelation functions with a central-peak width of 18 ps. The corresponding predicted hyperacuity (A75 from Altes 1989) is about 86 ns. The computer simulation of jitter discrimination trials predicts an acuity (At75 from Menne and Hackbarth 1986, Fig. 12) of about 50 to 55 ns for the same noise condition. Both predictions are based on a single echo's signal-to-noise ratio or detectability index, and both assume that fully coherent (phase-sensitive, using the crosscorrelation fine structure) processing of echoes is used. A semicoherent receiver would achieve substantially poorer performance under the same noise conditions. Table 1 summarizes predictions for both coherent and semicoherent reception of echoes at various detectability indices, including the 36-dB value representing the data collected here. RMS bandwidth is assumed to be 55 kHz (reciprocal of 18 Ixs), and centralized RMS bandwidth is assumed to be 13.9 kHz (Menne and Hackbarth 1986). In Table 1 there is general agreement between the two sets of predictions (ATs and A/75 ) across different detectability indices, while the actual performance of the bats (40 ns) falls slightly better than would be predicted for the case of single-echo reception.

D i s c u s s i o n

The original jittered-echo discrimination experiment with Eptesicusfuscus (Simmons 1979) produced two significant results. First, the acuity of jitter discrimination was smaller than 1 I~s, and, second, the shape of the jitter-discrimination curve approximated the half-wave- rectified crosscorrelation function of echoes. (The data for Bat # 1 and Bat # 2 were obtained in that original experiment.) Adding the data collected in the present experiment, the overall shapes of the basic (0 ~ phase) jitter discrimination curves for all 6 bats are alike (Fig. 6). The performance of the bats for jitter values in the region of 30 to 35 ps is consistently and significantly poorer than


for jitter values of 5 to 25 ~ts or 40 to 50 IXS. This peak in the error curve corresponds to the first side-peak in the crosscorrelation function of echoes (see below). Two of these bats were run with analog delay lines (Bat # 1 and # 2) and 4 bats were run with digital delay lines (Bat # 3 - # 6). The principles of operation of these devices are radically different; nevertheless, the bats perform as though only the electronically-set delay imposed on sounds passing through the simulator is responsible for discrimination. Specific properties associated with one type of delay device but not the other thus could not have provided unintended cues for the discrimination or caused the secondary error peak to appear at 30 to 35 lxs. The jitter discrimination curves evidently characterize the bat's echo-reception process with regard to echo delay in the context of the experiments rather than the properties of the specific apparatus chosen to produce the delay of echoes.

Eptesicus can perceive very small changes of less than 1 ~ts in echo delay. Under controlled conditions, with adequate practice and training, the bat can detect a shift in delay as small as 10 ns (Fig. 7). The repeatability of this same threshold estimate for different bats using both electronic delay lines and simple cable lengths strength- ens the conclusion that this is a reliable estimate of the bat's perceptual limitations under the conditions prevailing in these experiments. The original experiment (Sim- mons 1979) and two recent additional experiments (Menne et al. 1989; Moss and Schnitzler 1989) were able to measure jitter performance down to less than 1 ~ts, but the equipment was not intended to work in the range of tens of nanoseconds.

We presume that the extraordinarily small time change of 10 ns represents a hyperacuity for perception of echo delay (see Altes 1989), with useful acuity being less accurate than this value. Hole-depth and two-glint discrimination experiments suggest a normal acuity of the order of 1 IXS (Schmidt 1988; Simmons et al. 1974; Simmons et al. 1989), which in practice could be any- where in the range of several hundred nanoseconds to several microseconds. Experiments with barn-owls show a binaural temporal acuity of about 1 ~ts (Moiseff and Konishi 1981), while the electric fish, Eigenmannia, detects modulations of 0.4 Ixs or less in electric stimuli (Carr et al. 1986; Rose and Heiligenberg 1985), and in both instances the timing of neural discharges across a population of neurons actually conveys the crucial information from peripheral to central sites. On comparative grounds it is therefore perhaps not surprising that Eptesicus can perceive echo delay very accurately. The most conservative possibility is that bats, too, can process temporal stimulus information that is represented temporally in the nervous system to recover submicrosecond-sharp images along the axis of time. The value of 10 ns is extreme- ly small, however, and the critical question is whether spectral cues generated by interaction of stimulus echoes with extraneous echoes might have contributed to sharp- ening the bat's performance. The delay lines and electronic equipment did not introduce spectral artifacts (see Fig. 4), but artifacts could have been created as the electronically-returned echoes (al, az, or b) arrived at the

bat's ears in the company of echoes from various reflecting surfaces located near the bat on the platform. Routine sorts of control procedures related to spectral artifacts are described below, but what is wanted is some means of showing that the bat's perception of the delay of echoes depends upon the timing of neural discharges rather than upon physiological coding of the amplitude of echoes at some frequencies relative to others.

Echo crosscorrelation functions and jitter performance

Of greatest interest is whether the jitter discrimination data from the present experiments resemble the crosscorrelation functions of the echoes received by the bat. Our experiments, particularly the phase-shift procedure, were intended primarily to examine this question. Figure 17A shows spectrograms for sounds emitted by Bats # 3-6, and Fig. 17B shows durations of representative echolocation sounds recorded from all 6 bats during experimental trials. Bats # 3 and # 4 used sounds shorter than 2 ms, Bats # 1 and 6 used sounds as long as 2.5 ms, Bat # 2 used sounds as long as 3 to 3.5 ms, and Bat # 5 used sounds as long as 4 ms. Nevertheless, jitter discrimina-

A

B

I I I I I I I I I I I - 1 0 0 - 8 0 - 6 0 - 4 0 - 2 0 0 20 40 60 80 tO0

time (psec)

Fig. 19. Representative autocorrelation functions for (A) signals emitted by Bats ~ 3-~ 6, and for (B) 3 consecutive signals emitted by Bat 4~ 3 for jitter discrimination. Sampling rate is 2 MHz


don performance was not different in any respect. Figure 19A shows the autocorrelat ion functions of representative sounds from Bats # 3-6. Although the spectrograms of signals emitted by these bats differed, the autocorrelation functions are very much alike. The shapes of these functions in the region o f + 40 laS (the region most re- levant for the jitter data shown above) are similar. Figure 19B shows autocorrelation functions for 3 consecutively- emitted sounds recorded from Bat qr 3. These are even less variable than the functions shown for all 4 bats, primarily because the orientation of the bat's head did not change very much from one sound to the next. It is likely that the variability shown in Fig. 19A originates in different orientations of the bat's head relative to the recording microphone as much as in differences between the signals themselves. Interestingly, the autocorrelation functions in Fig. 19A are more nearly alike than are the spectrograms in Fig. 17A, indicating that the variations in the spectrograms may reflect a kind of surface description of the waveforms that does not adequately convey their underlying, deeper similarity - which the bats themselves can sense, judging from the similarity of the performance curves in Fig. 6. The broad features of the autocorrelation functions are relatively insensitive to the variations shown in the spectrograms, which consist mostly of differences in the distribution of the same frequencies over time rather than differences in the frequencies themselves. Figure 20A shows spectrograms of 7 consecutive signals recorded from one bat (Bat # 3) during the scanning stage of a trial to show the stability of the shape of the spectrogram when the bat is perform- ing a well-defined task and not moving very much relative to the microphone. Figure 20B shows the mean autocorrelation function of 50 signals emitted during several trials by Bat # 3 to give an indication of the variability o f the function over a set of signals actually

A -# T

~'~' 100 I ~r 50

0 ~--I 1 rnsec

B ._o

0 o 0

-ls -x~ T sd

I 1-1 I I I I I -75 -50 -25 0 25 50 75

-1 time (psec)

Fig. 20. (A) Spectrograms of 7 consecutively emitted signals recorded from Bat ~ 3 and (B) the mean (~ -4- 1 standard deviation) of the autocorrelation functions for 50 signals emitted by Bat 4~ 3 for jitter discrimination. Sampling rate in B is 250 kHz. Time scale in A refers to the individual spectrograms; the spacing between them is not related to actual repetition-rates used by the bats

Crosscorrelation Function

o~

- - ; o - ; o - ~ o - ~ o - ~ o 6 1'o io 3'0 go s'o Time (microseconds)

Jitter Performance

if J" e ./.~.~. .~.~.\.

~ % / " d

s - ; 0 - ; o - % - ~ o - i o 6 1'o 2'o 3'o g0 5'o Time (microseconds)

Fig. 21. Crosscorrelation function of an echo returning to the bat for comparison with the mean compound jitter discrimination curve for Bats ~ 3-~ 5. The jitter discrimination data are reflected around the zero origin of the time axis to form a symmetrical curve

used for a series of discrimination choices. Eptesicus is capable of emitting signals that have very similar features either in the time-frequency plane or in the summary statement of the time domain provided by the autocorrelation function. Furthermore, the autocorrelation functions for signals emitted by different bats are as much alike as the performance curves f rom the experiments.

The compound error curves derived by combining the 0 ~ and 180 ~ phase-shift data (see Figs. 9 and 10) are approximations to the point-spread function along the delay or range axis for the sonar of Eptesicus (Altes 1989). This is equivalent to the spatial impulse-response of the system, by analogy with the use of compound period histograms in auditory physiology (Kiang et al. 1965; Goblick and Pfeiffer 1969). The crosscorrelation function between emissions and echoes is the realization of the impulse-response for a particular emitted waveform. Figure 21 compares the crosscorrelation function of echoes reaching the bat for a representative signal (from Fig. 19B) with the mean compound jitter-discrimination curve for the 3 bats (Bat # 3, # 4, and # 5) that had side-peaks in their error curves at 35 ~ts. (Bat # 6 had its peak at 30 ~ts, so including it in the average would distort the resulting waveform by mixing sets of data having different inherent periodicities.) F rom Fig. 21, it is evident that the compound error curve is a good ap- proximation to the crosscorrelation function of echoes. Thus, both of the principal findings of the original jitter experiment are confirmed in our present data and extended to new observations.

A total of 4 recent experiments with Eptesicus have addressed whether the crosscorrelation function of

606 J.A. Simmons et al. : Jittered sonar echoes in echolocat ion

echoes might influence discrimination performance. In 2 of these, the procedure was an adaptation of two-choice echo-delay discrimination methods (Simmons 1973) in which model echoes were delivered to the bat. Time- reversing the model echo waveform (changing the downward FM sweeps to upward sweeps) drastically disrupts the accuracy of delay discrimination by bats (Masters and Jacobs 1989). Phase-shifting the model echo waveform displaces the apparent delay of echoes by an amount predicted from the crosscorrelation function of phase-shifted echoes (Masters 1989). These results are consistent with the appearance of the crosscorrelation function between signals and echoes in the original and present jitter data.

Two other recent experiments used the jittered-echo method itself; they did not obtain the crosscorrelation- like result shown in Fig. 21. In one of these experiments (Menne et al. 1989), the bats discriminated between jittering and nonjittering echoes presented in a two-choice task that was almost the same as the task used here. (One stimulus is presented on the left, the other stimulus on the right.) In the other experiment (Moss and Schnitzler 1989), jittering echoes were presented in a two-choice yes-no task. (The bat must go to one side if the stimulus is jittering and to the other side if it is not. Only one simulator channel exists - it presents jittering and nonjittering conditions in random order, and only the platform has left and right sides.) Both of these experiments yielded the same results for the basic jitter discrimination test - Eptesicus detected jitter values as small as 400 ns. The equipment used in these experiments could not make delay steps smaller than 400 ns, however, so the results cannot be taken as threshold estimates, just as upper bounds for the threshold. Figure 22 shows the data from the yes-no experiment (Moss and Schnitzler 1989) for delay steps of 5 Its from zero to 40 Its. The yes-no techni- que measures the percentage of hits made by each bat, and the numbers run from zero to 100% rather than from 50 to 100% as in two-choice experiments, so the numeri- cal scores are not strictly equivalent to the percentages of correct responses shown in Fig. 6. The curves in Fig. 22 show a mean performance that is relatively flat at about 85 to 90 % hits across different jitter values from

100 . . . . . -

~ 6o i- I [ o - - o Bat A ~ 40 i f * A - - ~ Bat B

J r m--m Bat C v - - . Bat D

20 i * - - * Bat E ] o o Bot F

0 i i i i i I i I i

0 5 10 15 20 25 30 35 40

Time (microseconds)

Fig. 22. Ji t ter discr iminat ion per formance (percentage hits) for 6 Eptes i cus in a yes-no parad igm that places echo b at the same delay as echo a z (Moss and Schnitzler 1989)

5 to 40 I~S. In contrast, the curves in Fig. 6 show about 95% correct responses over these same jitter values, except for the region of poorer performance around 30 to 35 Its. The ceiling of noisiness in the percentage-correct- responses data thus appears to be higher in Fig. 6 than in Fig. 22. Although the data in Fig. 22 support the observation that Eptesicus can perceive small changes in echo delay, they fail to show significantly poorer performance in the region around 30 to 35/as. A pronounced drop in performance at these time values is a salient feature of the curves in Fig. 6. The other recent jitter experiment (Menne et al. 1989) used a two-choice procedure that yields data extending from 50 to 100% correct responses that can be compared directly with our present data. The results of this other study (their Type 1 and Type 2 experiments) are much like the results of the yes-no experiment; performance is flat at about 90 % correct responses across the whole range of jitter values tested. The bats could discriminate small changes in echo delay, but no region of poorer performance appears at 30 to 35 laS.

In the original jitter discrimination experiment (Sim- mons 1979) and in our present experiments, the jittering echoes (al and a2) moved back and forth around a mean delay value (aav) that equalled the delay of the nonjittering or stationary echoes (b). That is, the delays of al and a2 in Fig. 1 were centered around the delay of b. In the recent yes-no version of the jitter experiment (Moss and Schnitzler 1989), the longer of the two jittering delays rather than the mean, equalled the delay of the nonjittering echoes. That is, a2 and b were the same in delay, with al being shorter. In the recent two-choice jitter discrimination experiment (Menne et al. 1989), the alignment of echoes again was different from that used in our present experiment. In this case, the shorter of the jittering delays, rather than the mean, equalled the delay of the nonjittering echoes. That is, al and b were at the same delay, with a2 being longer. These differences in the placement of the jittering and stationary echoes along the delay axis may have been responsible for the absence of a specific region of poorer performance around 30 to 35 ~ts in the yes-no results (Moss and Schnitzler 1989) and the other two-choice results (Menne et al. 1989). These other experiments placed the nonjittering echoes at the same delay as one of the two jittering echoes, creating a condition for masking from the trace of one echo to the other. Such masking is a real possibility; an interaction between images of jittering echoes is responsible for the presence of both the central peak and other peaks and nulls in the jitter discrimination curves. Data which actually demonstrate depressed jitter discrimination performance when the nonjittering echoes coincide in delay with either of the jittering echoes have been collected elsewhere (Fig. 4C in Simmons et al. 1990).

In view of the possibility that the alignment of the jittering and nonjittering echoes might account for the different results obtained in different jitter experiments, we emulated the alignments used in the other experiments and repeated the basic jitter experiment (0 ~ phase; jitter values from zero to 50 laS) using our equipment and procedures. When we used an off-center placement of

J.A. S immons et al. : Jittered sonar echoes in echolocation 607

100

~ 9o

go

7o 0 0

~ 6O

m 50

Bat #3

o--m a2= b I - - - - I ( ] C l V = b

100

0 90

n~

8O

70 3 ~ 8o

~ 50

B a t #5

~ - - o a2= b e - - - * Oav=b

r i i l i i r i i i i

0 5 10 15 20 25 30 35 40 45 50 Time (#.see)

Fig. 23. Graphs compar ing the per formance o f Bats # 3 and # 5 in the two-choice jitter discrimination task with echo b equal in delay to echo a 2 or to the mean delay of a 1 and a 2. Each data point is 40-60 trials

jittering and nonjittering echoes (al =b or a 2 = b), the performance of two tested bats (Bat ~ 3 and ~ 5) was the same as that obtained in the jitter experiments done elsewhere. Figure 23 shows the basic jitter performance of these two bats for both off-center and on-center placement of stimuli along the delay axis. The off-center curve (a2 = b) shows a pattern similar to that observed in the yes-no experiment (Moss and Schnitzler 1989; see Fig. 22 above), with no region of poorer performance at 30 to 35 las. The off-center curves in Fig. 23 are also identical to the results of the other two-choice jitter experiment (Menne et al. 1989; their Fig. 7). A comparison of Figs. 6, 22, and 23 reveals that the placement of echo b relative to a~ and a2 indeed is critical for whether the specific region of poorer performance at 30 to 35 las occurs in the data. The region of better performance between zero and 25 gs, which is prominent in Fig. 6, is rendered poorer in the off-center conditions in Fig. 23, leading to the specific region of poorer performance at 30 to 35 Ixs being partially obscured. In addition, the region of poorer performance at 30 to 35 ixs may be abolished, quite apart from its being obscured by a general rise in the number of errors made at other jitter values by the bats. In other words, the 30-35 las side-peak in the data is indistinct in the results of the other jitter experiments due to a general rise in the percentage of errors at jitter values besides 30-35 gs and perhaps also partly due to masking of whatever echo information causes the side-peak to occur by the coincidence of a2 and b. The representation of echo information leading to the side-peak may not be as robust as the representation of information related to

other aspects of jitter discrimination, such as acuity smaller than 1 ~ts. The results of the present experiment (Fig. 6) appear now to be reconciled with the differing results of the yes-no experiment (Moss and Schnitzler 1989) and the other two-choice experiment (Menne et al. 1989). The side-peak at 30 to 35 Ixs in our jitter data is real, it just was lost in the other experiments as a consequence of the choice of delays for the jittering and nonjittering echoes. By replicating their delay alignments in our procedure (Fig. 23), we find the same difference between the two sets of data; the disappearance of the side-peak at 30 to 35 ~ts in the data from off-center echo aligments reveals a possible masking effect that may be specific to the side-peak itself. Masking between the nonjittering echoes and one of the pair of jittering echoes would be an instance of clutter interference, and the disappearance of the 35-~ts side-peak in the data may be a consequence of adaptation of receiver properties to suppress such clutter and facilitate detection of the other of the pair of jittering echoes (Altes 1971).

One other strategy for proving that the bat's behavior mirrors the crosscorrelation function of echoes is to change the periodicity inherent in the crosscorrelation function and see whether the bat's behavior changes accordingly. The experimental condition in which echoes were filtered to a narrow band around 55 kHz is just such a test. The crosscorrelation function ordinarily has a single prominent side-peak located about 30 ~ts away from the main peak (see Figs. 19-21), but this spacing is changed in the 55 kHz condition to about 18 ~ts. Further- more, more of the side-peaks at multiples of 18 lxs become prominent as a result of the restricted bandwidth of the 55-kHz condition. We take this particular result in Fig. 12 to confirm that the jitter discrimination performance of Eptesicus indeed does have approximately the same shape as the crosscorrelation function of echoes. The next question to be considered concerns whether the bat's behavior truly reflects the image of a target in the bat's perception or is merely a consequence of spectral cues generated by artifacts.

Echo signal-to-noise ratio and fine acuity

The results obtained by introducing noise into the stimuli returned to the bat demonstrate the robust nature of the bat's fine nanosecond-sized acuity. At an echo detectability index of 36 riB, Eptesicus can perceive a jitter of 40 ns, which is slightly better than the estimates in Table 1 of 50-55 ns (At75) or 86 ns (ATs) to be expected for an ideal, coherent receiver operating on one echo for each delay. The question remains as to whether the bat can enhance the echo signal-to-noise ratio by integrating energy across several echoes that might be delivered for any one delay value. Usually, several echoes returned to the bat for each delay, and it is difficult to settle on a single number to represent the minimum that the bat needs to achieve such fine acuity in the noise. The bats' 40-ns performance is roughly 10 times better than the 348-400 ns acuity predicted for a semicoherent receiver (Table 1), and a sufficient number of echoes to achieve


a factor of 10 improvement in semicoherent performance just did not occur. Probably the best conclusion is that Eptesicus performs the jitter task approximately as an ideal coherent receiver, with some allowance for imper- fect integration of energy across multiple echoes over the time course required to present one or more echoes at any given delay value during a whole trial. This conclusion is reinforced by the data shown in Fig. 21, which documents the coherent character of the image more directly.

Acoustic cues for jitter discrimination

The principal concern for the interpretation of the jittered-echo data is the nature of the acoustic cues actually used by the bat to perceive small changes in the delay of the echoes. The bat could perceive the arrival-time of each echo separately, observing the jitter in terms of changes in apparent echo delay from one emission to the next. In this case, the timing of each echo initially would be encoded by the timing of neural discharges evoked by each frequency in the echo (Bodenhamer and Pollak 1981; Bodenhamer et al. 1979; Pollak et al. 1977; Sim- mons and Kick 1984; Suga 1970, 1988), and echo delay ultimately is likely to be represented by auditory mechanisms associated with the transformation of neural- discharge timing into a place display of target range (O'Neill and Suga 1982; Suga and Horikawa 1986; Suga and O'Neill 1979; Suga 1988; Sullivan 1982; Wong and Shannon 1988). The bat's capacity to perceive the jitter would depend upon its ability first to produce an image of each echo along a perceptual delay axis and then to detect a shift in the location of this image from one echo to the next. The shape of the jitter performance curve over a wide range (zero to 50 gs) would thus reflect the shape of the image itself (Altes 1989; Simmons and Stein 1980). The sharpness of the performance curve in the immediate vicinity of zero would reflect the acuity of the subsequent process of detecting shifts from one image to the next. Sharpness in a hyperacuity task is approximated by the quantity Avs (Aires 1989). These two aspects of performance correspond roughly to the two stages of signal-processing associated with an ideal pulse- compression sonar receiver - first, forming the crosscorrelation function and, second, forming the a posteriori distribution (Woodward 1964). The data reported above indicate that Eptesicus perceives delay (range) images having a crosscorrelation-like fine structure on a scale of tens of microseconds and additionally can detect shifts in these images with an accuracy in the region of 10 ns. Furthermore, the bat's performance of 40 ns at an echo detectability index of 36 dB is approximately what would be expected for the delay accuracy of an ideal receiver.

As an alternative to perceiving echo delay as such, the bat could be perceiving changes in the amplitude or the spectrum of echoes that are correlated with different magnitudes of electronic delay introduced by the simulator's delay lines. That is, changes in echo arrival-time could interact with some other feature of the stimulus configuration to produce amplitude changes or spectral cues that might have gone unnoticed to everyone but the

bats. These cues could occur as artifacts concealed within the delay-dependent transfer function of the target simulator or they could be created by overlap of stimulus echoes from the simulator with the bat's emissions or with extraneous echoes from any reflecting surfaces located near the bat in the experiment. The crucial echo cues would then be changes in the overall amplitude of echoes from one jittered delay to the other, or changes in amplitude at particular frequencies, rather than delay itself. For example, if the stimulus echoes were to overlap with additional echoes reflected from some object located near the bat, the jittered delays could be transformed into small shifts in the frequency of notches in the spectrum of the combined echoes reaching the bat's ears. The bat might be perceiving movements of these spectral notches along the frequency scale as a means of determining which side of the left-right simulator was delivering the jittered echoes. There are two ways to deal with this alternative set of non-delay or spectral cues. One way is essentially passive - to measure the signals emitted by the bat and the echoes returned from the loudspeakers and from any objects located near by as a way of determining whether such cues might be present. For instance, if the delay lines exhibit different spectral properties at different delays, the magnitude of this effect could easily be observed as part of the procedure for calibrating the simulator. The other way is active - to demonstrate some property of the bat's performance that is compatible only with direct perception of delay or, alternatively, only with perception of artifactual cues that are different from delay. This requires exploiting some physiological dis- tinction between the representations of temporal and spectral information.

Passive controls

Amplitude or spectral cues could creep into the stimuli reaching the bat's ears because the delay lines or their associated circuits introduce delay-related changes in signal strength or spectrum as part of the process of retarding signals. Detecting such cues is part of the process of calibrating the simulator. For analog delay lines or cables, a progressive increase in the attenuation of high frequencies indeed occurs as the amount of delay increases (the length of the delay line increases). How- ever, this attenuation only encroaches upon the frequencies below 100 kHz used by Eptesicus when the total delay achieved by analog means is much greater than that actually obtained here by analog means. Analog devices were used here as supplements to digital delay to minimize spectral artifacts that would have been present if the total delay of about 3 ms had been obtained entirely in an analog mode. For example, the maximum delay obtained from analog lines for the data in Fig. 7 was only 60 ns. For the larger delays generated entirely digitally, amplitude and spectral cues were at most 1 bit out of the 12-bit range of the devices because transfer functions computed with 12-bit data showed no gain or spectral changes related to the amount of delay selected (see Fig. 4).


As part of the calibration process, the total delay imposed by the simulator was measured directly using a digital counter in a time-interval mode; an accuracy of 3 ns could be achieved by this method. Within this accuracy, the nominal electronic delays were indeed achieved as intended with the appropriate delay devices, whether this was done by setting switches on analog delay lines, by setting address bits on digital delay lines, or by cutting RG-58U coaxial cables to the desired lengths. The transfer function of the simulator was measured for actual stimulus conditions by passing 1-ms FM signals (10 to 100 kHz) through the system, with various cable lengths or delay settings on the delay lines. Figure 4 shows transfer functions for jitter conditions of 0, 5, 10, 15, and 20 ns. The transmission-line properties of the simulator can be summarized as a frequency-dependent amplitude variation of about 0.05 dB that is independent of the delay settings, even with analog delays in the critical nanosecond range, and an accumulating, frequency-dependent phase shift corresponding to the time delay imposed by the delay lines. (The phase shifts in Fig. 4 are referred to the absolute delay of echo al; 3 . 2 7 5 x 106 ns+3 ns as a zero-phase reference.) The transfer functions also show fluctuations across frequency that originate in the sampling rate (250 kHz) and the window used on the data (Hamming window) rather than in the delay lines themselves. Thus, there is no evidence for the existence of amplitude or spectral cues within the simulator that the bat could use to detect small amounts of jitter in delay apart from delay itself.

Amplitude or spectral artifacts could also appear after the signal emerges from the delay line and loudspeaker as a consequence of interaction between the electronic echo traveling back to the bat's ears and extraneous echoes reflected by parts of the apparatus or the room. The primary source of incidental echoes that might overlap with stimulus echoes is the part of the apparatus that juts out in front of the bat on the Y-shaped platform. This consists of the microphones and loudspeakers, together with their supports. From their distances and locations at the ends of the two arms of the platform (see Fig. 1), the microphones and loudspeakers should reflect echoes starting with a delay of about 1 ms. Figure 24 shows the waveform and spectrogram of all sounds - including both stimulus and extraneous echoes - arriving at the location of the bat's ears during a period of over 9 ms following the emission of a 2-ms imitation Eptesicus sonar signal (102 dB SPL peak-to-peak) by an electrostatic loudspeaker placed approximately at the site of the bat's mouth. The test sounds were digitally generated (16 bits; 1-MHz sampling rate) using an IBM- PC-AT computer and an R. C. Electronics 200 Series waveform synthesizer. These sounds were recorded with an uncovered B and K Model 4135 condenser microphone placed at the location of the bat's ears, amplified with a Princeton Applied Research Corporation Model 113 low-noise preamplifier, and digitized in an IBM PC-AT computer using an R. C. Electronics Model ISC-16 data acquisition board (500 kHz sampling rate; 12-bit accuracy). Because the B and K microphone's intrinsic noise made it possible to record only the stron-

A

kHz

f c~ 80 B

z 6o

N=512 C~ 40 ~ :

I I I 0 4 8 ms

apparatus echoes~l stimulus echo I I floor echo I

emission I I 1 m e t e r

TIME (ms)

Fig. 24. Waveforms (A) and spectrograms (B) of 2-ms digitally- synthesized Eptesicus emission, apparatus echoes, stimulus echo, and floor echo recorded at the location of the bat's head during discrimination trials. To improve the signal-to-noise ratio, 512 signals were averaged. The stimulus echo is isolated from extraneous overlapping echoes over a dynamic range of 30-35 dB

gest of the extraneous echoes, 512 artificial bat sounds were broadcast and the recorded signals were averaged on-line to improve the signal-to-noise ratio. From Fig. 24A, the microphone directly picked up the 2-ms bat sounds at great strength (emissions), together with (in succession) reflections from the arms of the platform, the microphones, and the loudspeakers (apparatus echoes), the electronic echo (b) passing through the simulator (stimulus echo), and weaker echoes identified as coming from the floor located 110 to 115 cm below the platform. (The attenuation of simulated echoes was set to yield a simulator gain of - 2 7 dB, which was the value used for Bat 4# 4.) Similar recordings of echoes were made with a bat in its place on the platform and emitting sounds at the targets. These real echoes were identical to what appears in the average shown in Fig. 24A except that the noise level of the recordings obscured the low-level reflections around 3 ms and around 7-8 ms on the time scale. For sonar emissions with a 2-ms duration, the stimulus echoes are isolated in time from any extraneous echoes having an amplitude within about 30-35 dB of their own amplitude. The spectrograms (Fig. 24B) for the sounds returning to the bat's location show the direct pick-up of the emission mixed with multiple, overlapping echoes from the apparatus (having a spectrum with numerous peaks and notches), the stimulus echo (having a clean spectrum that reflects only the loudspeaker frequency response - see Fig. 5), and then a weaker echo from the floor.

Overlap of the raw waveform of the stimulus echo with the emission could only occur for emissions longer than 3.2 ms. Overlap with strong apparatus echoes could only occur if the bat's sound is longer than about 2.1 to 2.2 ms. From Fig. 24, for a 2-ms emission, the apparatus echoes were essentially finished by about 3 ms of delay, and stimulus echoes began no earlier than 3.2 ms following emissions. Two of the bats (4#3 and 4#4) appeared


not to emit sounds longer than 1.5 to 2 ms during jitter discrimination trials (see Fig. 17B). These bats did not encounter overlap of stimulus echoes with extraneous echoes in any of our recordings. Two other bats (4~ 1 and # 4) appeared to abruptly cut off the duration of the sounds they emitted at durations of about 2.2 to 2.5 ms. These bats encountered several hundred microseconds of overlap for those trials where any of the sounds were longer than about 2.1 ms. However, Bat # 4 was recorded as making sounds shorter than 2 to 2.1 ms on trials where it still responded correctly, so the small overlap that did occur for many of its longer sounds was not a requirement for its performance. Not enough recordings were available for Bat ~ 1 to draw this same conclusion. The remaining two bats (41:2 and ~ 5) emitted sounds as long as 3 to 4 ms on every recorded trial and encountered up to 1-2 ms of overlap for echoes of these sounds. Other extraneous echoes were substantially weaker than the echoes from the platform, microphones, and loudspeakers, and they occurred after, not before, the stimulus echoes. The nearest parts of the walls, ceiling, and floor of the room were roughly 0.5 m beyond the simulated range of the test targets (about 56 cm). From Fig. 24, the time that elapsed between onset of the stimulus echoes and the beginning of any echoes from the surfaces of the room, however attenuated they may be by the convoluted foam panels, was over 3 ms, which rules out overlap of stimulus echoes with these incidental echoes for emissions shorter than 3 ms. The sounds of Bats # 4 and # 5 often were longer than this, so they encountered an appreciable overlap of stimulus echoes with later, weaker incidental echoes. For the other bats, though, no overlap with later echoes was possible. Thus, the performance of the bats in Figs. 6 and 7 could not have depended on overlap of echoes, with the resulting generation of amplitude or spectral cues, because performance was not affected by whether overlap occurred. For bats emitting longer-duration sounds, the introduction of what might seem to be potentially potent spectral cues led to no discernable change in performance compared to that of the bats whose sounds were short enough to avoid overlap.

Could the bats thought to use short sounds actually be using sounds longer than 2.1 ms without this fact being noticed? In the experiments, the emissions are picked up with Briiel and Kjaer Model 4138 microphones, which are insensitive enough that the electrical signal-to-noise ratio of the recorded sounds coming out of the microphones and going into the simulator is the limiting step for the signal-to-noise ratio of the echoes delivered to the bats as stimuli, unless the noise genera- tors were operating. The dynamic range of the remainder of the simulator is wide enough that no further noise is added. Several special recordings were made with one of the bats observed to use only short sounds (Bat # 3) to determine whether the durations measured for sounds passing through the simulator might be truncated significantly by the noise floor established by the B and K microphones. A QMC Model SM-1 microphone was used in parallel with the B and K microphones to record the sounds of this bat, and the durations of the same

sounds were then estimated from both raw signal waveforms and from spectrograms recorded through both microphones. The QMC microphone is roughly 28 dB more sensitive than the B and K microphone and thus ought to produce longer estimates of signal duration if a pronounced truncating effect is indeed operating on the durations described above. The signals of Bat 4~ 3 were not significantly longer in duration in the recordings made with the QMC as compared to the B and K microphones. Evidently the rising and falling portions of the envelopes of the sounds are so steep as they approach the margins of the sound that the lower noise-level of the QMC microphone does not yield much longer sounds. It thus appears impossible for overlap of stimulus echoes with emissions or incidental echoes to be a source of spectral cues for Bat ~ 3 to perceive small changes in echo delay in the jitter discrimination task. If some of the bats did use added spectral cues, their performance should have improved, or at least changed, but it did not.

The remaining source of overlap between stimulus echoes and extraneous echoes is the bat's own peripheral auditory system. If the bat's emissions or the echoes from the microphones and speakers were retained mechanical- ly by the inner ear for a few milliseconds, they might then mix with stimulus echoes to create delay-dependent spectral cues (see Menne et al. 1989). The Organ of Corti is an active mechanical system, and ringing or other post- stimulus vibration of the basilar membrane is well-known to occur in animals, including bats that emit CF signals and have a cochlea specialized for CF reception (Henson et al. 1987; K6ssl and Vater 1985). These CF bats have auditory filters with very sharp tuning (high Q values), so extended periods of ringing are to be expected. FM bats have not been shown to possess such sharply-tuned inner-ear auditory filters: the degree of tuning that has been observed is consistent with time-constants of only a few hundred microseconds (Simmons et al. 1989). Ob- servations of cochlear microphonic responses are needed to determine whether longer retention of vibration occurs in FM bats (some preliminary experiments suggest that it does not). We think that the bat does not make critical use of spectral cues generated by overlap of echoes with trace vibrations of the basilar membrane because the apparent delay of echoes changes according to the amplitude-latency trading ratio for NI and N4 responses to FM sounds (see below). In addition, the durations of the emissions were different from one bat to another, which ought to alter the details of overlap between stimulus echoes and intrinsic cochlear vibrations, yet performance was unchanged.

Active controls

Measurements of the bat's sounds and the properties of the simulator seem to rule out any extraneous spectral cues being available to at least some of the bats for perceiving the small changes in echo delay demonstrated in the jitter discrimination experiments. Overall, the hypothesis that overlap-generated spectral artifacts are the basis for jitter performance cannot explain the in-

J.A. S immons et al. : Jittered sonar echoes in echolocation 611

variance of performance with the presence or absence of echo overlap for different bats, so the hypothesis would seem to be incorrect. However, the changes in delay perceived by the bat are so incredibly small that one would like some further assurance of a fundamentally temporal basis for the bat's performance. A more active approach to identifying the auditory basis for jitter discrimination involves taking steps to nullify the usefulness of spectral amplitude cues, to deliberately introduce spectral cues to see if performance improves, or identifying some crucial feature of performance that is not consistent with a spectral hypothesis. The constancy of the bats' performance in spite of the variability of signal duration falls into the latter category.

Clutter interference experiments have measured the degree of overlap between echoes that is required for Eptesicus to perceive two separate echoes as starting to merge together - that is, to overlap in an auditory sense (Simmons et al. 1988, 1989). These results indicate that the time separation must be smaller than about 300-350 gs for interactions to occur between echoes. The form of the data specifically indicates that spectral cues are generated when echoes fall closer together than this figure but that no such spectral interaction occurs for time separations that are much greater (Simmons et al. 1989). The integration-time of the sonar receiver determines the echo separation required for spectral effects to appear (Beuter 1980), and the measured value of 300-350 gs is far shorter than the prevailing separations of about 2 ms or more between stimulus and extraneous echoes in the present jitter experiments. The jittering echoes and the echoes from the microphones and loudspeakers therefore fall well outside the bat's time window for auditory mixing of echo waveforms due to overlap, even if the raw waveforms are seen on an oscilloscope to overlap for some of the bats. The equivalent integration-time for the oscilloscope is the duration of the horizontal axis of the display, which is not a good analogy with auditory integration-time. The flaw in arguing that the bat necessarily experiences spectral cues when the raw waveforms overlap is that the bat receives sounds through the band-pass filters of the inner ear before encoding them as neural discharges, while an oscilloscope does not. The duration of the mechanical excitation in each filter channel should be the true duration of each echo for purposes of generating interference cues in the spectrum. Unless overlap is shorter than 300-350 gs, spectral cues might not occur at all. Thus, it is possible that spectral cues were unavail- able to the bats even when the raw echo waveforms themselves overlapped, as must have occurred regularly for Bats ~ 2 and ~ 5. If the behaviorally-measured clutter effect (Simmons et al. 1989) is a good guide to the generation of spectral cues, then all 6 bats would be expected to perform the same in the jitter experiments, even though they emitted sounds of different durations, because no spectral cues were in fact present.

If the bat uses spectral cues to detect the small incre- ments in delay for the jittering target, the effective auditory stimulus (the stimulus echo in Fig. 24 when it reaches the bat's inner ear) must consist of local increases or decreases in stimulus amplitude at specific frequencies

(spectral peaks and notches). The amount of change in amplitude and perhaps the specific frequencies of the spectral effects must be correlated with the delay nominally being presented for each stimulus condition in order to yield the observed performance. In other words, the size of the spectral cues available at 20 ns jitter must be larger than the cues available at 10 ns jitter. Spectral cues that are not correlated with the amount of jitter cannot account for the curves shown in Figs. 6 and 7. Uncorrelated spectral cues would merely raise the overall level of performance at all jitter values, rather than yield a curve that slopes down to chance performance (50 % errors) at zero jitter. This observation suggests several control procedures for spectral cues.

First, there is a source of variation in the echo spectrum that is not correlated in any way with the amount of jitter, and it is overwhelmingly greater than the minimal changes in echo amplitude or spectral level actually observed in sounds reaching the bat's location when emissions are shorter than 2.1 ms. (For longer emissions, overlap of the raw waveform of echoes of course occurs, and larger spectral effects are seen if one assumes that an oscilloscope display is appropriate for judging auditory overlap. These cues do not appear, however, if the echoes are filtered into frequency bands similar to the auditory tuning curves of FM bats before being analyzed [Sim- mons et al. 1989].) Figure 4 shows the variability of about 0.05 dB electronically introduced into signals passing through the simulator. Any additional spectral cues, correlated with jitter, added to the stimuli after being broadcast from one of the loudspeakers must be smaller than 0.1 to 0.2 dB across frequencies, because no effects of this size were observed in the echoes shown in Fig. 24. When the bat's sounds are recorded by the left and right microphones of the simulator, the inherent noise of the microphone preamplifiers adds variability to the spectrum that is uncorrelated with jitter and independent of any of the stimulus conditions. This variability is passed through the simulator to appear in the echoes delivered to the bat. We recorded 8 separate sets of 10 echoes presented during 20 ns trials with Bat # 3 (which always used sounds shorter than 2.1 ms) and found the standard deviation of echo amplitude (8 sets of N= 10) to vary from 2 to over 10 dB from 23 to 90 kHz as a consequence of the simulator's microphone noise. Spectral notches and peaks of 10 to 25 dB were introduced by the microphone noise into all 80 individual echoes at random frequencies. This uncorrelated spectral variability superimposed on the potential correlated spectral variability of less than 0.1 to 0.2 dB made detection of any correlated variability impossible, even with in-phase ensemble averaging of echoes used to produce Fig. 24. To detect jitters of 10 to 20 ns using spectral cues while emitting sounds shorter than 2.1 ms, the bat would have to detect spectral variations of a fraction of a decibel that are correlated with the amount of jitter in the presence of much larger random variations from one echo to the next. For reception and in-phase averaging of as many as 10 echoes at each delay (al, a2 or b), this appears impossible, and the bats frequently received fewer than 10 echoes for each delay (see Fig. 16). In the experiments in which electronic noise


was deliberately introduced to lower the signal-to-noise ratio of echoes (Fig. 18), even the microphone noise was undetectable, and echo spectral cues were obscured further. The standard deviation of echo spectral level increased to 10 to 25 dB across frequencies from 23 to 90 kHz, again in a manner independent of the amount of jitter or other aspects of the bat's behavior during trials. Under these conditions, using longer sounds that produced oscilloscope overlap of echoes and correlated spectral peaks and notches based on the artificially long integration-time of that display, the noise-introduced spectral effects also masked the peaks and notches created by interference between overlapping echoes. Yet, the two bats tested were able to achieve discrimination of 40 ns jitter. We conclude that artifactual spectral cues could not have played a decisive role in jitter discrimination.

The second approach to controlling for the bat's use of spectral cues correlated with the amount of jitter is to deliberately modify the spectrum of one of the jittering echoes (al) relative to the other jittering echo (a2) and to the nonjittering echoes (b). This added spectral cue should make all jitter values readily discriminable, resulting in a flat performance curve at a level better than chance with no sharp decline around zero jitter. In 3 different experiments with Bats ~ 3 and ~ 4, we added a second echo component to echo a~ at a time separation of 10, 20 or 30 ixs. These separations are much smaller than the integration-time for echo reception, so the two components of al should interfere to create well-defined, broad spectral notches that are 30 to 35 dB deep at a frequency of 50 kHz (10 Ixs separation), at frequencies of 25 and 75 kHz (20 ~ts separation), or at frequencies of 17, 50, and 83 kHz (30 ~ts separation). When presented with these strong spectral cues that distinguish echo a2 from the other echoes, the bats did not show an overall improvement in performance at all jitter values, as would be expected from the use of spectral cues. Instead, the bats made significantly more errors at the particular jitter values of 10, 20 or 30 Ixs, as would be expected if the spectral cues built into al were transformed into an estimate of the time separation of the two components of at prior to being expressed in the image of what is now a two-glint target (Simmons et al. 1990). Thus, not only did the two bats fail to exploit spectral cues to sharpen jitter discrimination performance, but they perceived each of these examples of spectral cues as an event at the appropriate location along the time axis of the performance curve (Simmons, unpublished).

Temporal cues for jitter discrimination

The assumption behind our experiments is that the delay of each echo (al, a2, b) is separately represented by the bat's neural mechanisms for determining target range, and that the small changes in delay which constitute the jittering stimulus are perceived as changes in the apparent absolute range of the simulated target. The image evoked by one echo must be explicitly compared with the image evoked by the next echo for the bat to detect the

jitter, implying some sort of memory or persistence of the image from one sonar emission to the next. When jittering echoes coincide in delay, the resulting decrease in discrimination performance reflects interference between one image and the other because they both presumably activate the same subset of delay-representing neurons. Furthermore, when a particular delay difference (35 Its in Fig. 6, for example) produces a decline in performance, some residual activation of the same subset of neurons must occur, too. As a consequence, a graph of the bat's performance (percentage correct or percentage errors) yields a kind of profile of the distribution of the trace of the image along the perceived range axis. This profile specifically represents recovery of information that has been stored: In effect, it is an outline of the image itself. Using the jitter procedure, the image profile, or the point- spread function for range, thus can be determined directly (Altes 1989).

If the bat indeed encodes each echo in terms of delay and then feeds this encoded representation into its range- determining mechanism, the apparent range of the target should be susceptible to latency shifts induced indirectly by changes in echo amplitude as well as directly by changes in delay. The time-of-occurrence of neural discharges marking successive individual frequencies in the FM sweeps of emissions and echoes (Bodenhamer and Pollak 1981 ; Simmons and Kick 1984) provides the input to the neural display of target range (Suga 1988), and these discharge-times can be perturbed by changing the amplitude of one of the two jittering echoes relative to the other. Measurements of evoked potentials in Ep- tesicus at the level of the auditory nerve (N1) and lateral lemniscus (N4) show these latency shifts in most on- responding neurons to be - 13 to - 18 tas/dB (Simmons et al. 1990). Figures 13 to 15 demonstrate that Eptesicus experiences this same magnitude of shift in the apparent distance to the simulated targets in the jitter experiment. If artifactual spectral cues are responsible for jitter discrimination, the bat's performance should not shift along the time axis as echo amplitude changes but should remain locked to the timing conditions that create the specific pattern of spectral peaks and notches serving as the artifactual cues. We conclude that the bat achieves its performance in the jitter discrimination task by using the time-of-occurrence of neural discharges to represent echo delay prior to formation of the images that it perceives. These images can be shifted along the bat's psychological range axis as either echo delay or echo amplitude changes. That is, the images track the timing of neural discharges by whichever acoustic means that timing is altered. Interestingly, neurons in the lateral lemniscus have been shown to encode the timing of frequencies in FM sweeps with an accuracy (Covey and Casseday 1989) that might be sharp enough to explain the presence of phase information as instantaneous frequency in the images along the delay axis (Simmons and Grinnell 1988).

We believe that amplitude-latency trading of range effectively eliminates delay-related changes in echo spectra as critical cues for jitter discrimination. Eptesicus evidently perceives the absolute delay of echoes from the timing of neural discharges. Even though the bat may


move its head in range by 0.3 to 1 mm from one sound to the next, which is enough to disrupt the location of its head relative to the loudspeakers on the fine scale of 0.002 mm implied by 10-ns discrimination, the bat seems able to overcome this diff• Perhaps the bat senses the movements of its head and takes them into account in judging distance. Or perhaps the bat detects the jitter in al/a 2 by referring the delay of the stimulus echoes to the delay of echoes reflected several milliseconds earlier from the microphones and loudspeakers (see Fig. 24). The delay of these earlier echoes will be disrupted by approximately the same amount as the delay of the stimulus echoes when the bat moves its head. In either case the timing of neural discharges evoked by the stimulus echoes would convey the necessary information to the bat, which is consistent with amplitude-latency trading.

The jitter procedure not only yields data about the detectability of small time shifts, it yields a picture of the dependence of detection on the size of the jitter interval, which corresponds to the shape of the image iself (Altes 1989). The location in frequency of spectral peaks and notches does not change with overall echo amplitude, but the bat's perception of echo delay does change with echo amplitude, matching the magnitude of change associated with latency shifts. Furthermore, the bat's perception of differences in target range when spectral cues are deliberately introduced (created when echoes overlap as they arrive at the bat's ears) is independent of relative echo amplitude (Simmons et al. 1990). This experimental dis- sociation of temporal and spectral cues for image structure shows that both types of cues exist and that both cues have an effect on what the bat perceives along the range axis, but it also shows that absolute target range is extracted from temporal cues, with spectral cues serving to modify the shape of the image as a whole around its time-based core.

The crosscorrelation-like side-peaks in the image (Fig. 21) are an expression of echo spectral composition, but in the domain of time rather than frequency. It would be misleading to say that the bat computes the crosscorrelation images without reference to spectral information, for surely the bat's auditory representation of FM echoes depends upon having frequency-tuned neural channels that convey the frequency axis of the spectrogram of echoes (see Simmons et al. 1989, 1990). What counts, however, is that the crucial computations take place on the time-of-occurrence of events in these neural channels rather than on their frequencies per se (Sim- mons 1980). Otherwise, perceived range would not be subject to amplitude-latency trading. Recent experiments establish that Eptesicus indeed uses an explicitly spectral representation for some components of the image, even though they ultimately are expressed in the image as values of time-separation rather than as values of frequency (Simmons et al. 1990). The same process (first representing spectral features of echoes in the frequency domain but then transforming them into the time domain) could well be involved in the presence of side- peaks in the jitter performance curves at 35 gs (they are moved to 20 and 40 ps in the 55-kHz-filtered condition shown in Fig. 12). However, the latency-shift effect (Figs.

13 to 15) carries the side-peaks, and even the entire 0 ~ and 180" structure of the crosscorrelation function, along with the main peak when echo amplitude is changed. For this to occur, the bat must truly perceive the phase of echoes relative to emissions on a time scale defined by the latency of neural discharges (see below).

The amplitude-latency trading experiments illustrate a problem that can occur in jitter discrimination tasks. The extreme narrowness of the main peak in the performance curve makes it difficult to carry out any jitter experiment in which this peak is shifted to a new, only approximately-known, location. In our experiments, this peak is only about 20 ns wide and can easily be missed if time steps no smaller than several microseconds are used. In previous jitter experiments with 4-45 ~ echo phase-shifts (Menne et al. 1989), the main peak was expected to shift from zero to about 7 gs on the horizontal scale, but no jitter values actually were sampled between 4 and 8 gs. Since even the data from those experiments indicated that the main peak must be less than 1 gs wide, to say nothing of its astonishing narrowness in Fig. 7, the possibility that time and phase might trade off was not tested in those experiments through omission of jitter values more closely clustered around 7 gs.

Acoustic images and echo signal-processing

The results of the jitter discrimination experiments with 0 ~ and 180 ~ echo phase-shifts (Fig. 10) demonstrate that Eptesicus incorporates the phase of echoes relative to emissions in the images it perceives. This phase information somehow is associated with the time-of-occurrence of neural discharges representing echo delay, as is shown by the amplitude-latency trading effect on the whole image. That is, the experimentally-induced shift in the apparent delay of echoes that corresponds to the neural amplitude- latency trading ratio carries the phase information along with it (Figs. 13-15). However, the magnitude of latency shifts that can be induced (200 ~s and more; Simmons et al. 1990) is greater than the duration of the rising slope of individual cycles at even the lowest frequencies in echoes (about 22 gs at 23 kHz). Thus, neural discharges probably are not triggered by individual cycles of echoes in a phase-locked manner but instead are triggered by the much longer duration of the envelope of the excitation delivered to first-order neurons through the band-pass filters of the hair-cells (Simmons et al. 1989). In other words, in a conventional sense the phase of signals is destroyed when emission and echo waveforms are encoded by neural discharges. Yet, the fine structure of the crosscorrelation function evidently is conveyed up the auditory pathways by the time-of-occurrence of neural discharges to the site at which crosscorrelation (or its equivalent) occurs. These neural discharges are triggered by the timing of individual frequencies in the FM sweeps of emissions and echoes (Bodenhamer and Pollak 1981; Bodenhamer et al. 1979; Pollak et al. 1977). The information collectively represented by the frequency to which ascending auditory neurons are tuned and the timing of their discharges amounts to the instantaneous frequency


of each emission and echo. Such a representation con- stitutes a neural spectrogram in the form of an array of discharges traveling upward through the lateral lemniscus and inferior colliculus (Simmons and Kick 1984). The width of the envelope of excitation delivered to auditory-nerve fibers by hair-cells presumably establishes the integration-time for the numerous impulse-like wavelets (spectral slices in the frequency domain) making up the spectrogram (Simmons et al. 1989). Besides de- stroying conventional phase information, this integration- time apportions the information required to reconstruct the target's image into both the time and frequency dimensions of the spectrogram (Altes 1980, 1984), requir- ing the bat to explicitly operate on both temporal and spectral auditory representations to recover the image (Simmons et al. 1990). The critical aspect of neural coding that recovers the apparently lost phase information may lie in the integration of time and frequency dimensions into the image the bat ultimately perceives.

Earlier experiments on target ranging by bats had indicated that the formation of the bat's acoustic images must depend on some process whereby the instantaneous-frequency patterns in emissions and echoes can be transformed into envelopes of crosscorrelation functions (Simmons 1973). The envelope of the crosscorrelation function of emissions and echoes indeed can be recovered from a spectrogram representation. In fact, an echo spectrogram contains all information about the echo except for a constant phase shift; multiplying the echo by - 1 does not affect its spectrogram. It follows that an echo spectrogram contains information that can be used to obtain crosscorrelation fine structure as well as envelope, except for the effect of a constant phase shift (Altes 1980, 1981). (The crosscorrelation function corresponds to the unsquared ambiguity function or time- frequency correlation function [Van Trees 1971] evaluated along the time axis for zero Doppler shift.) However, the displacement of the fine structure of the crosscorrelation function in the 180 ~ phase-shift experiment goes beyond what can be extracted from instantaneous frequencies alone. To reconstruct the whole crosscorrelation function requires knowledge of instantaneous frequencies in emissions and echoes plus knowledge of a constant phase-shift between them. A spectrogram correlation hypothesis (Altes 1980, 1981) is sufficient to account for the envelope of the function appearing in the jitter data, but not the fine structure after the phase shift. To account for the jitter results in their entirety, the spectrogram representation residing in the bat's auditory system must be augmented with phase information. If each instantaneous-frequency point or cell in the spectrogram carries its own phase information along with its time-of- occurrence, the crosscorrelation function then might be recovered through a spectrogram-correlation process. The bat actually perceives a crosscorrelation-like image for a band-limited portion of the echo around 55 kHz (Fig. 12), which suggests that the requisite phase information indeed may be conveyed locally with each frequency. The basis for such a spectrogram-with-phase representation could be in the sharpness of timing of neural discharges; if the variability inherent in the regis-

tration of instantaneous frequency were to be smaller than the period at some frequencies, phase-like information would be incorporated along with the timing of the discharges and might emerge in the image as a consequence of delay computations across frequencies (Sim- mons 1980), and the requisite timing accuracy may be present in Eptesicus (Covey and Casseday 1989).

The finding that Eptesieus perceives echo phase is a direct challenge to conventional views of the auditory neural representation of acoustic stimuli, in which phase information is not thought to be preserved in the neural code at frequencies above a few kHz. The fine delay hyperacuity of about 10 ns is also a challenge. How can the nervous system determine the time-of-occurrence of echoes with an accuracy several orders of magnitude smaller than the known or suspected accuracy of the timing of neural discharges in single nerve cells (see Carr et al. 1986)? Among other issues, it seems certain that the effective accuracy of the neural representation of timing is better than has previously been thought possible. It seems certain, too, that fine delay acuity arises from neural processing in a substantial population of cells, very likely cells organized in a topographic delay map (Suga 1988) that has specific connections with cells organized tonotopically as well (Simmons et al. 1990). One particularly intriguing implication of the fine 10-ns acuity is that the readout from the range map may involve not only the location of activated neurons but also the de- tailed time structure of the ensemble activity across the map. Averaging of neural activity - that is, interpolation within the delay map - seems unlikely to improve accuracy enough to get from the hundreds of microseconds exhibited by single cells to the 10-ns accuracy of the bat as a whole. Some specific processes must select which neural discharges are reliable prior to being averaged, and the relative timing of discharges could be a basis for such selection. Thus, the bat's fine acuity may arise in part because the stimulus dimension being processed - time - happens also to be one of the response dimensions used to read images from neural activity.

There is also the question of why the bat should bother keeping track of echo delay with such fine acuity. It is true that a range-profile basis for shape perception by FM bats (Simmons and Chen 1989; Simmons et al. 1990) would require more accurate target-range registration than the 1-2 cm required merely to capture prey, but surely not to within 10 ns or so. Perhaps the bat uses this fine acuity for bilateral target-range and range disparity computations related to horizontal localization of targets from binaural differences in echo arrival-time (Simmons et al. 1983). The relatively small binaural time differences to be expected from the small size of the bat's head would be within the limits of useful perception even if the bat's practical acuity were 100 times worse than 10 ns. Binau- ral intensity differences could contribute greatly to range- disparity localization by magnifying interaural time differences through binaural amplitude-latency trading (Pollak 1988). In principle, all that would be required is separate determination of the distance to each reflecting glint from the left and the right ears, combined with neural connections between the left and the right images


to detect range disparities. The demons t r a t ed convers ion of spectral to tempora l i n f o r m a t i o n a b o u t target range and range profile (S immons et al. 1990) might be the basis for us ing range dispari ty to perceive the hor izonta l direct ion no t jus t of whole targets bu t the direct ion of par ts of targets, too. I f Eptesicus could achieve this exten- sion of range-axis i n fo rma t ion in to direct ional localiza- t ion, it might t ruly see objects as emergent images in three d imensions .

Acknowledgements. This research has been supported by Office of Naval Research Contract No. N00014-86-K-0401, by NIH Grant No. DC00511, by Grant No. BNS 83~)2144 and prior grants from the National Science Foundation, by NIMH Research Scientist Development Award No. MH00521, by a grant from the System Development Foundation, and by a University Research Instru- mentation Grant from the Department of Defense. We thank R. R. Capranica, J. H. Casseday, E. Covey, D. R. Griffin, A. D. Grinnell, J. Mcllwain, D. Menne, G. D. Pollak, H.-U. Schnitzler, A. M. Simmons, N. Suga, W. E. Sullivan and 3 reviewers for their con- structive suggestions. These experiments were conceived in response to challenges posed by J. J. O'Hare, H. Hawkins, and others at the Office of Naval Research in a series of meetings held in 1985 and 1986. We are grateful for having the opportunity to address them.

References

Altes RA (1971) Suppression of radar clutter and multipath effects for wide-band signals. IEEE Trans Information Theory IT-17:344-346

Altes RA (1976) Sonar for generalized target description and its similarity to animal echolocation systems. J Acoust Soc Am 59 : 97-105

Altes RA (1980) Detection, estimation, and classification with spectrograms. J Acoust Soc Am 67:1232-1246

Altes RA (1981) Echo phase perception in bat sonar? J Acoust Soc Am 69:1232-1246

Altes RA (1984) Texture analysis with spectrograms. IEEE Trans Sonics Ultrasonics SU-31:407-417

Altes RA (1989) Ubiquity of hyperacuity. J Acoust Soc Am 85: 943-952

Beuter KJ (I 980) A new concept of echo evaluation in the auditory system of bats. In: Busnel RG, Fish JF (eds) Animal sonar systems. Plenum Press, New York, pp 747-761

Bodenhamer RD, Pollak GD (1981) Time and frequency domain processing in the inferior colliculus of echolocating bats. Hear- ing Res 5:317-355

Bodenhamer RD, Pollak GD, Marsh DS (1979) Coding of the fine frequency information by echoranging neurons in the inferior colliculus of the Mexican free-tailed bat. Brain Res 171 : 530-535

Busnel RG, Fish JF (eds) (1980) Animal sonar systems. Plenum Press, New York

Cahlander DA (1964) Echolocation with wide-band waveforms: Bat sonar signals. MIT Lincoln Laboratory Technical Report No. 271, Lexington, MA

Carr CE, Heiligenberg W, Rose GJ (1986) A time-comparison circuit in the electric fish midbrain. I. Behavior and physiology. J Neurosci 6:107-119

Covey E, Casseday JH (1986) Connectional basis for frequency representation in the nuclei of the lateral lemniscus of the bat Eptesicusfuscus. J Neurosci 6:2926-2940

Covey E, Casseday JH (1989) Parallel monaural pathways to the midbrain in an echolocating bat: The ventral lateral lemniscal complex (VLL). Soc Neurosci Abstr 15 : 746

Goblick TJ, Pfeiffer RR (1969) Time domain measurements in cochlear nonlinearities using combination click stimuli. J Acoust Soc Am 46:924-938

Griffin DR (1958) Listening in the Dark. Yale Univ. Press, New Haven, CT (2nd ed., Cornell Univ. Press, Ithaca, NY, 1986)

Habersetzer J, Vogler B (1983) Discrimination of surface-structured targets by the echolocating bat Myotis myotis during flight. J Comp Physiol 152 : 275-282

Hartley DJ (1989) The effect of atmospheric sound absorption on signal bandwidth and energy and some consequences for bat echolocation. J Acoust Soc Am 85:1338-1347

Henson OW Jr, Bishop AL, Keating AW, Kobler JB, Henson MM, Wilson BS, Hansen RC (1987) Biosonar imaging of insects by Pteronotus p. parnellii, the mustached bat. National Geographic Res 3:82-101

Kawasaki M, Margoliash D, Suga N (1988) Delay-tuned combination-sensitive neurons in the auditory cortex of the vocalizing mustached bat. J Neurophysiol 59:623-635

Kiang NYS, Watanabe T, Thomas EC, Clark LF (1965) Discharge patterns of single fibers in the cat's auditory nerve, Res Monog 35, MIT Press, Cambridge, MA

Kick SA (1982) Target detection by the echolocating bat, Eptesicus fuscus. J Comp Physiol 145:431-435

Kick SA, Simmons JA (1984) Automatic gain control in the bat's sonar reciever and the neuroethology of echolocation. J Neuro- sci 4: 2705-2737

Krssl M, Vater M (1985) Evoked acoustic emissions and cochlear microphonics in the mustache bat, Pteronotus parnellii. Hearing Res 19:157-170

Lawrence BD, Simmons JA (1982) Measurements of atmospheric attenuation at ultrasonic frequencies and the significance for echolocation by bats. J Acoust Soc Am 71 : 585-590

Lazzaro J, Mead C (1989) Silicon modeling of pitch perception. Proc Natl Acad Sci USA 86:9597-9601

Licklider JCR (1951) A duplex theory of pitch perception. Ex- perientia 7:128-134

Masters WM (1989) Tradeoff between echo crosscorrelation-peak delay and time delay in target ranging by big brown bats. Soc Neurosci Abstr 15 : 512

Masters WM, Jacobs SC (1989) Target detection and range resolution by the big brown bat (Eptesicusfuscus) using normal and time-reversed model echoes. J Comp Physiol A 166:65-73

McCue JJG (1966) Aural pulse compression by bats and humans. J Acoust Soc Am 40:545-548

Menne D, Hackbarth H (1986) Accuracy of distance measurement in the bat Eptesicus fuscus: Theoretical aspects and computer simulations. J Acoust Soc Am 79:386-397

Menne D, Kaipf I, Wagner I, Ostwald J, Schnitzler HU (1989) Range estimation by echolocation in the bat Eptesicusfuscus: trading of phase versus time cues. J Acoust Soc Am 85 : 2642-2650

Moiseff A, Konishi M (1981) Neuronal and behavioral sensitivity to binaural time differences in the owl. J Neurosci 1 : 40--48

Moss CF, Schnitzler HU (1989) Accuracy of target ranging in echolocating bats: acoustic information processing. J Comp Physiol A 165:383-393

Nachtigall P, Moore PWB (eds) (1988) Animal sonar: Processes and performance. Plenum Press, New York

O'Neill WE, Suga N (1982) Neural encoding of target range and its representation in the auditory cortex of the mustached bat. J Neurosci 47:225-255

Pollak GD (1988) Time is traded for intensity in the bat's auditory system. Hearing Res 36:107-124

Pollak GD, Marsh DS, Bodenhamer R, Souther A (1977) Charac- teristics of phasic-on neurons in the inferior colliculus of unanes- thetized bats with observations related to mechanisms for echoranging. J Neurophysiol 40:926-941

Pye JD (1986) Sonar signals as clues to system performance. Acus- tica 61 : 166-175

Rose GJ, Heiligenberg W (1985) Temporal hyperacuity in the electric sense of fish. Nature 318:178-180

Schmidt S (1988) Evidence for a spectral basis of texture perception in bat sonar. Nature 331:617-619

Schnitzler HU, Henson OW Jr (1980) Performance of airborne

616 J.A. Simmons et al. : -littered sonar echoes in echolocation

animal sonar systems: I. Microchiroptera. In: Busnel RG, JF Fish (eds) Animal sonar systems. Plenum Press, New York, pp 109-181

Schnitzler HU, Menne D, Hackbarth H (1985) Range determination by measuring time delay in echolocating bats. In: Michelsen A (ed) Time resolution in auditory systems. Springer, Berlin Heidelberg New York, pp 180-204

Simmons JA (1969) Depth perception by sonar in the bat Eptesicus fuscus. PhD Thesis, Princeton Univ, Princeton, N,l

Simmons JA (1973) The resolution of target range by echolocating bats. J Acoust Soc Am 54:157-173

Simmons JA (1979) Perception of echo phase information in bat sonar. Science 207:1336-1338

Simmons ,lA (1980) The processing of sonar echoes by bats. In: Busnel RG, Fish ,iF (eds) Animal sonar systems. Plenum Press, New York, pp 695-714

Simmons JA (1989) A view of the world through the bat's ear: the formation of acoustic images in echolocation. Cognition 33 : 155-199

Simmons ,lA, Chen L (1989) The acoustic basis for target discrimination by FM echolocating bats. J Acoust Soc Am 86:1333-1350

Simmons JA, Grinnell AD (1988) The performance of echolocation: Acoustic images perceived by echolocating bats. In: Nachtigall P, Moore PWB (eds) Animal sonar: Processes and performance. New York, Plenum Press, pp 353-385

Simmons ,lA, Kick SA (1984) Physiological mechanisms for spatial filtering and image enhancement in the sonar of bats. Annu Rev Physiol 46:599-614

Simmons JA, Stein RA (1980) Acoustic imaging in bat sonar: Echolocation signals and the evolution of echolocation. J Comp Physiol 135 : 61-81

Simmons JA, Vernon JA (1971) Echolocation: Discrimination of targets by the bat Eptesicusfuscus. J Exp Zool 176:315-328

Simmons JA, Lavender WA, Lavender BA, Doroshow DA, Kiefer SW, Livingston R, Scallet AC, Crowley DE (1974) Target structure and echo spectral discrimination by echolocating bats. Science 186:1130-1132

Simmons JA, Howell D.l, Suga N (1975) Information content of bat sonar echoes. Am Sci 63:204-215

Simmons JA, Fenton M B, Ferguson WR, Jutting M, Palin J (1979) Apparatus for research on animal ultrasonic signals. Royal Ontario Museum Publications in Life Sciences, Toronto

Simmons JA, Kick SA, Lawrence BD, Hale C, Bard C, Escudie' B (1983) Acuity of horizontal angle discrimination by the echolocating bat, Eptesicusfuscus. ,l Comp Physiol 153:321-330

Simmons JA, Kick SA, Moffat AJM, Masters WM, Kon D (1988) Clutter interference along the target range axis in the echolocating bat, Eptesicusfuscus. J Acoust Soc Am 84:551-559

Simmons JA, Freedman EG, Stevenson SB, Chen L, Wohlgenant TJ (1989) Clutter interference and the integration time of echoes in the echolocating bat, Eptesicus fuscus. J Acoust Soc Am 86:1318-1332

Simmons JA, Moss CF, Ferragamo M (1990) Convergence of temporal and spectral information into acoustic images of complex sonar targets perceived by the echolocating bat, Eptesicus fuscus. J Comp Physiol A 166:449-470

Strother GK (1961) Note on the possible use of ultrasonic pulse compression by bats. ,l Acoust Soc Am 33:696-697

Suga N (1970) Echo-ranging neurons in the inferior colliculus of bats. Science 170:449-452

Suga N (1988) Auditory neuroethology and speech processing: Complex-sound processing by combination-sensitive neurons. In: Edelman GM, Gall WE, Cowan WM (eds) Auditory function. Wiley, New York, pp 679-720

Suga N, Horikawa J (1986) Multiple time axes for representation of echo delays in the auditory cortex of the mustached bat. J Neurophysiol 55: 776-806

Suga N, O'Neill WE (1979) Neural axis representing target range in the auditory cortex of the mustache bat. Science 206:351-353

Sullivan WE (1982) Neural representation of target distance in auditory cortex of the echolocating bat Myotis lucifugus. J Neurophysiol 48: 1011-1032

Van Bergeijk WA (1964) Sonic pulse compression in bats and people: A comment. J Acoust Soc Am 36:594-597

Van Trees HL (1971) Detection, estimation, and modulation theory, part III. Wiley, New York

Wever EG (1949) Theory of hearing. Wiley, New York Wong D, Shannon SL (1988) Functional zones in the auditory

cortex of the echolocating bat, Myotis lucifugus. Brain Res 453 : 349-352

Woodward PM (1964) Probability and information theory, with applications to radar, 2nd Edit. Pergamon, New York

Zbinden K (1988) Harmonic structure of bat echolocation signals. In: Nachtigall PE, Moore PWB (eds) Animal sonar: processes and performance. Plenum, New York, pp 581-587

Documents

Discrimination of jittered sonar echoes by the ... · ic maps in the bat's auditory system presumably provide a topographic display of echo spectral features-in par- ticular, the