Waves

The bat that lowers its call to hold its echo still

A horseshoe bat calls at one long, pure note near 83 kilohertz, and flying towards a wall at five metres a second it hears the echo come back two and a half kilohertz higher. So it lowers its call, by exactly the factor its flight raises the echo, and the echo arrives at the same pitch whatever the bat's speed. The reason is its ear, tuned so sharply that a change of speed of a third of a metre a second would carry an uncorrected echo out of the band it hears best. With the still world parked on one note, anything that moves stands out — and a moth's beating wings write themselves into the echo as a comb of sidebands.

Assumes: The shift a mirror gives twice · The note that changes on approach, and the two ways of getting it

Most bats that hunt by sound use short calls that sweep downward in frequency, a chirp of a few milliseconds, and listen for how long the echo takes to return. The horseshoe bats do something else. They call with a long, nearly pure tone — tens of milliseconds at a single frequency, with a short sweep tacked on at the end — and they listen not mainly to when the echo returns but to its pitch.

In the greater horseshoe bat that tone is near 83 kilohertz, varying a little between individuals and populations. In 1968 it was found that a flying horseshoe bat does not keep that tone fixed. As it flies faster towards a wall, it calls lower, and it lowers its call by just the amount needed to make the echo come back at a fixed pitch. The behaviour is called Doppler shift compensation, and it turns the Doppler effect, which for a radar or a speed gun is the measurement, into a nuisance to be removed. Why a bat should want to remove it, and what it gains by doing so, is a question about waves as much as about bats.

Two Doppler shifts, one on the way out and one on the way back

A source moving towards a stationary listener compresses the waves it sends ahead of it, and the listener hears a higher frequency, f(c)/(c−v)f(c)/(c - v). A listener moving towards a stationary source meets the waves more often than they arrive, and hears f(c+v)/cf(c + v)/c. The note that changes on approach separated these two cases, which are not symmetric for sound, because sound has a medium and the medium picks out a frame.

An echo involves both. The bat is a moving source on the way out, so the wall receives the call raised by c/(c−v)c/(c - v); the wall reflects what it receives, and the bat is then a moving listener on the way back, which raises it again by (c+v)/c(c + v)/c. The product is

fecho=fcall c+vc−v,f_\text{echo} = f_\text{call}\,\frac{c + v}{c - v},

the two-way shift that the shift a mirror gives twice found for radar. For a bat at 5 metres a second in air where sound travels at 343, the factor is 1.0296, and an 83-kilohertz call comes back 2.46 kilohertz higher. Even a slow flight of half a metre a second raises it by 240 hertz.

The bat lowers its call by exactly what its flight adds. Frequency against a bat's speed towards a stationary reflector. A call held at 83.2 kHz would come back raised by the two-way Doppler shift (dashed), 2.46 kHz at 5 m/s. A horseshoe bat instead lowers its call (the falling curve) by 2.39 kHz at 5 m/s, the same factor (c − v)/(c + v) the flight multiplies by, and the echo (solid) arrives at 83.2 kHz at every speed. Round numbers for a greater horseshoe bat.
Fig. 1 Frequency against a bat’s speed towards a stationary reflector. A call held at 83.2 kHz would come back raised by the two-way Doppler shift (dashed), 2.46 kHz at 5 m/s. A horseshoe bat instead lowers its call (the falling curve) by 2.39 kHz at 5 m/s, the same factor (c−v)/(c+v)(c - v)/(c + v) the flight multiplies by, and the echo (solid) arrives at 83.2 kHz at every speed. Round numbers for a greater horseshoe bat.

The compensation is exact in principle: the bat calls at fref(c−v)/(c+v)f_\text{ref}(c - v)/(c + v), the flight multiplies that by (c+v)/(c−v)(c + v)/(c - v), and the echo arrives at freff_\text{ref} whatever vv is. The bat cannot compute this; it works by feedback, adjusting each call according to where the previous echo landed, a few calls a tenth of a second. Measured in flight and on a swinging pendulum that carries a bat towards a target, the echo is held within a few tenths of a per cent of the reference — a few hundred hertz at 83 kilohertz, against a shift of thousands — and the reference sits a little above the frequency the bat uses when it is hanging still, by about a hundred and fifty to two hundred hertz.

An ear tuned too sharply to tolerate its own speed

The reason for all this is in the bat’s ear. The horseshoe bat’s cochlea devotes a disproportionate length of its basilar membrane, and its auditory brain a disproportionate number of neurons, to a narrow band of frequencies around the reference: an acoustic fovea, analogous to the patch of the retina packed with cones. Neurons tuned there are exceptionally sharp. A common measure of sharpness is Q10Q_{10}, the centre frequency divided by the width of the tuning curve ten decibels down from its tip; for ordinary mammalian hearing it is of order five to ten, and for the horseshoe bat’s foveal neurons values in the hundreds have been reported.

An ear tuned too sharply to tolerate its own speed. The sensitivity of a horseshoe bat's hearing near its reference frequency, drawn as a single resonance with a quality factor of 300 measured 10 dB down: 277 Hz wide at 10 dB down (shaded). The bat's own call at 83.0 kHz falls 13 dB below the peak even in this gentle one-resonance model, and real tuning curves fall more steeply on the low side, so the bat hears its own call far more faintly than its echo. Uncorrected echoes at 0.5, 1 and 2 m/s land at 83.44, 83.69, 84.18 kHz; a speed change of 0.29 m/s is enough to move an echo out of the band. Without compensation the ear would lose its echoes at walking pace.
Fig. 2 The sensitivity of a horseshoe bat’s hearing near its reference frequency, drawn as a single resonance of Q10=300Q_{10} = 300: 277 Hz wide at 10 dB down (shaded). The bat’s own call at 83.0 kHz falls 13 dB below the peak even in this gentle model, and real tuning curves fall more steeply on the low side. Uncorrected echoes at 0.5, 1 and 2 m/s land at 83.44, 83.69 and 84.18 kHz; a change of speed of 0.29 m/s is enough to move an echo out of the band.

A filter of Q10=300Q_{10} = 300 at 83.2 kilohertz is 277 hertz wide. The Doppler shift corresponding to half that width is a speed of c Δf/2fc\,\Delta f/2f, 0.29 metres a second. A bat whose ear is that sharp cannot afford to let its flight speed decide where its echoes land, because a change of speed of a third of a metre a second — a fraction of a wingbeat’s worth — would move an uncorrected echo out of the band it hears best. The compensation exists because the hearing is sharp, and the hearing can afford to be sharp because the compensation exists. Each is useless without the other.

The arrangement has a second benefit, visible in the same figure. The bat’s own call, at its resting frequency or below, falls on the low skirt of its own tuning, where the ear is much less sensitive. The call is loud — over a hundred decibels at a few centimetres — and the echo is faint, and both arrive at the same time, because a long call is still going out when the echo of a nearby object returns. A bat that heard its call and its echo equally well would be deafened by itself. By placing the echo in the fovea and the call just below it, the horseshoe bat separates them in frequency, the way a radio separates two stations, and can listen to a whisper while shouting.

That is a filter doing a job that elsewhere is done by timing. The speeds a pulsed radar folds into one dealt with a radar that cannot listen while it transmits and must alternate; a continuous-wave radar listens all the time and separates its transmission from its echo by the Doppler shift. Horseshoe bats are, in that sense, continuous-wave radars that have moved their own transmitter out of their receiver’s band.

A call that overlaps its own echo, and a filter that takes time to answer

The numbers make the separation by frequency necessary rather than merely clever. Sound covers 17 metres in 50 milliseconds, so the echo from anything closer than 8.6 metres — half that distance, since the sound has to go and return — starts arriving before the call has finished. For a bat hunting in a hedge, that is every echo it cares about. And it cannot hear far in any case: at 83 kilohertz air absorbs sound at a rate of order two or three decibels a metre, the steep rise with frequency that the distance that takes the treble out traced, so the echo of a moth is lost in a few metres. A horseshoe bat’s whole working world lies inside the distance at which call and echo overlap.

A sharp filter has a cost of its own. The narrower a resonance, the longer it takes to respond to a change, by the relation the width that is a lifetime established: a filter 277 hertz wide needs about a millisecond or more to build up or die away. That is fast enough to follow a moth’s wingbeat at fifty or a hundred hertz, which changes the echo every ten or twenty milliseconds, and too slow to follow the microsecond structure of a sweep. The fovea is built for exactly the timescale of the thing it is meant to detect, and the bat’s long tone gives it time to settle.

What a stationary world looks like after compensation

Once everything stationary returns at one frequency, the echoes of leaves, branches and walls pile up on a single line in the spectrum, however many of them there are and however far away. Anything that moves relative to them is the only thing left that can appear off the line.

A flying insect is such a thing, and its wings especially. A moth beating its wings fifty times a second has surfaces moving towards and away from the bat at metres a second, and the echo from those surfaces is Doppler-shifted back and forth fifty times a second. In the language of radio this is phase modulation: the path length to a wing changes by twice its displacement, the echo’s phase by 4πd/λ4\pi d/\lambda, and a sinusoidal modulation of phase with index β\beta spreads a pure tone into a comb of lines at multiples of the modulation frequency, with powers Jn(β)2J_n(\beta)^2.

A moth's wings write themselves as sidebands. The spectrum of the echo a horseshoe bat receives from a moth beating its wings at 50 Hz with a 3 mm swing along the line of sight, when the bat's compensation has put everything stationary at the reference frequency. The still world — foliage, walls, the moth's body — returns one line at the reference. The wings' moving surfaces phase-modulate their share of the echo by β = 4πd/λ = 9.1 radians, spreading it into lines 50 Hz apart reaching about ±457 Hz. Only something that flutters makes a pattern off the line.
Fig. 3 The spectrum of the echo a horseshoe bat receives from a moth beating its wings at 50 Hz with a 3 mm swing along the line of sight, after compensation has put everything stationary at the reference. The still world — foliage, walls, the moth’s body — returns one line. The wings phase-modulate their share of the echo by β=4πd/λ=9.1\beta = 4\pi d/\lambda = 9.1 radians, spreading it into lines 50 Hz apart reaching about ±457 Hz.

At 83 kilohertz the wavelength in air is 4.1 millimetres, so a wing swinging three millimetres along the line of sight changes the echo’s phase by nine radians, and the echo’s wing component spreads over about ±450 hertz in lines fifty hertz apart. The comb’s spacing gives the wingbeat rate, and its width the speed of the wing. Experiments with horseshoe bats show that they detect and attack fluttering targets and ignore still ones of the same size, and that they can discriminate insects by the pattern of these modulations — what is sometimes called acoustic glints, the bright flashes of echo when a wing is momentarily square to the beam.

The trick only works because compensation has cleared the background. Without it, the echoes of the leaves around a moth would be spread across a band set by the bat’s own changing speed and the angles to each leaf, and the moth’s sidebands would sit inside that clutter. With it, the clutter collapses to one line and the insect stands beside it. Radar engineers call the same idea a moving-target indicator, and the shift a mirror gives twice described the airborne radars that cancel the ground’s echo to see an aircraft against it; a horseshoe bat does it in a hedge, with a larynx and a cochlea.

Only a long call can be a speedometer

None of this is available to a bat with a short call. To measure a frequency to a precision Δf\Delta f takes a time of about 1/Δf1/\Delta f: a tone that lasts only a millisecond has a spectrum a kilohertz wide however pure it is meant to be. Sharpness has to be paid for made this a theorem about Fourier transforms before it became a principle of quantum mechanics, and it applies to an echo exactly as written.

Only a long call can be a speedometer. The smallest difference in speed a call can resolve through the Doppler shift, c/(2fT), against the call's length T, for calls at 83 kHz and 40 kHz. A frequency can be measured no more finely than one over the time spent measuring it. A horseshoe bat's 50 ms call resolves 4.1 cm/s; a 2 ms frequency sweep of the kind many other bats use resolves only 2.1 m/s, about the whole flight speed of an insect, and bats that call that way use their calls for range instead.
Fig. 4 The smallest difference in speed a call can resolve through the Doppler shift, c/(2fT)c/(2fT), against the call’s length TT, for calls at 83 kHz and 40 kHz. A horseshoe bat’s 50 ms call resolves 4.1 cm/s; a 2 ms frequency sweep of the kind many other bats use resolves only 2.1 m/s, about the whole flight speed of an insect, and bats that call that way use their calls for range instead.

A horseshoe bat’s call of fifty milliseconds can in principle resolve frequency differences of twenty hertz, or speed differences of four centimetres a second, which is what lets it read wingbeats as fine structure. A bat using two-millisecond sweeps from 40 kilohertz resolves only about two metres a second through the Doppler shift, which is useless for the purpose. Such bats get something else from their calls instead. A frequency sweep of bandwidth BB can be compressed on reception into a pulse of length about 1/B1/B, the operation the long pulse that arrives as a spike described in radar, and that gives a range resolution of about c/2Bc/2B — a few millimetres for a sweep tens of kilohertz wide.

The two kinds of bat are therefore making opposite choices on the same trade. A long pure tone is sharp in frequency and blurred in time: a speedometer and a flutter detector that cannot range well. A short sweep is sharp in time and blurred in frequency: a rangefinder that cannot measure speed. The horseshoe bats’ short sweep at the end of each call is a nod to the other side of the trade, and is thought to be used for ranging. No single call can be both sharp in time and sharp in frequency, and the anatomy of each family’s ear follows the choice its call has made.

The correction runs one way

Doppler shift compensation was found, in the experiments that established it, to work much better in one direction than the other. When the echo rises, as it does in forward flight towards anything, the bat lowers its call. When the echo falls, as it would if the bat were drawing away from a reflector, the call is not raised above its resting frequency by a matching amount.

The correction runs one way. Call and echo frequency against the closing speed between a horseshoe bat and what it hears, if the bat lowers its call for echoes that rise, as experiments find it does, and does not raise its call above the resting 83.0 kHz for echoes that fall. Approaching, the echo is held at the reference; drawing apart, it drops below the call, to 81.09 kHz at 4 m/s, outside the tuned band. Flight is almost always forward, so the shifts a bat meets are nearly all upward.
Fig. 5 Call and echo frequency against the closing speed between a horseshoe bat and what it hears, if the bat lowers its call for echoes that rise and does not raise it above the resting 83.0 kHz for echoes that fall. Approaching, the echo is held at the reference; drawing apart, it drops below the call, to 81.09 kHz at 4 m/s, outside the tuned band.

For a flying animal this is almost costless, because flight is nearly always forward: everything the bat is approaching returns a rising echo, and the things it is leaving behind are behind it, outside the beam of its call. The asymmetry has a second reading. The resting frequency is the top of the bat’s range of calls, and the reference sits just above it, so the bat’s own call can never land in its fovea; the system has been built so that its own voice is always on the deaf side of its hearing. A compensation that worked both ways would have to move the call up through the reference, and on the way it would deafen the ear it was meant to serve.

The same trick in a laboratory

The same arrangement — move the source so that what comes back lands on a fixed narrow detector — appears in places that have nothing to do with bats. Laser cooling of atoms tunes a laser slightly below an atomic resonance so that only atoms moving towards the beam see it Doppler-shifted into resonance, the effect the Doppler shift with nothing coming closer set in a wider context. A spectroscopist who wants to see the narrow line of an atom beneath its Doppler-broadened outline selects atoms by their velocity in a similar way. And the loudspeaker that goes round spun a source to make a Doppler shift that varied in time, which is what a moth’s wing does to a horseshoe bat’s echo.

The surprising part is not that an animal uses the Doppler effect but which half it uses. A traffic radar wants the Doppler shift and discards the echo of the stationary world. The bat removes the Doppler shift of the stationary world so that it can attend to the part of the echo that varies, and its own speed, which the shift encodes, is not what it is after. The quantity a physicist would call the signal is, to the bat, the background.

What these figures cannot show

The numbers are round values for one species: a reference near 83.2 kilohertz, a resting call 200 hertz below it, a filter of Q10=300Q_{10} = 300 drawn as a single symmetric resonance, a moth with a single sinusoidal wing motion. Real foveal tuning curves are asymmetric and have steep flanks and broad tails; real insects have two pairs of wings, bodies that also move and wingbeats that are not sinusoidal; real echoes from foliage are spread by the leaves’ own motion in wind. The figures show the principle and its scale, not a measured bat.

The Doppler formulas are exact for sound in still air but leave out wind, which moves the medium and so shifts both the call and its echo; and they treat the reflector as stationary or as a single moving surface. The compensation itself is described here as if it were instantaneous. It is not: the bat adjusts on the basis of echoes already received, and a bat accelerating rapidly — on landing, say — compensates with a lag. How the neural control loop is built, and how it handles several echoes at different distances arriving at slightly different frequencies, is the subject of continuing work in auditory neuroscience, and none of it is visible in a frequency plot.

Finally, the essay’s explanation of why compensation exists is a physical one — a sharply tuned receiver requires a stable input — and it is the explanation biologists give, but the order in which the sharp ear and the compensation evolved is not something the physics can settle.

Still open: what the bat does with several echoes at once

A horseshoe bat in a hedge receives echoes from leaves, twigs and insects at many distances and many angles, and the Doppler shift of each depends on the angle between the bat’s flight and the direction to it, as cos⁡θ\cos\theta. Compensation can hold only one of them exactly at the reference — presumably the strongest echo from straight ahead — while the others land slightly below it, spread by angle. How the bat chooses which echo to compensate for, whether its filter’s width is matched to the spread of angles in its beam, and how much of its prey detection depends on this geometry rather than on wingbeat sidebands alone, are not settled by a model of a single wall and a single moth.

A horseshoe bat calling near 83 kHz hears the echo of a stationary wall raised by the two-way factor (c+v)/(c−v)(c + v)/(c - v), 2.46 kHz at 5 m/s, and lowers its call by exactly that factor so that the echo always returns at one reference frequency — because its hearing is tuned there with a Q10Q_{10} of order 300, a band only 277 Hz wide that a change of speed of 0.29 m/s would carry an uncorrected echo out of; with the still world parked on one line, a moth’s wings stand out as a comb of sidebands ±450 Hz wide, readable only because a 50 ms call resolves speeds to 4 cm/s where a 2 ms sweep resolves only 2 m/s. The bat removes the Doppler shift a radar would keep, and keeps what the radar would discard.

Part 11 of 11

This essay is one argument about Doppler. The others:

The objects named here

The third axis, after the field and the reading path: the things themselves, and every essay that touches each one.

Auditory filterDoppler effectEcholocationFrequency resolutionPhase modulationSideband