USING BAT AUDITORY MODELLING FOR OBJECT

DISCRIMINATION AND ECHO SEPARATION TASKS

Dragana Nikolić, Timos Papadopoulos and Robert Allen

Institute of Sound and Vibration Research, University of Southampton, University Road, Southampton, U.K.

Keywords: Bat echolocation, Acoustics modelling, Echo separation, Object discrimination.

Abstract: This paper proposes a computational echolocation model that can be used for discrimination of the specific

target object and reconstruction of the acoustic information from multiple overlapping echoes returning

back from that target. An acoustic model for estimation of the backscattering impulse responses of a rigid

disk has been developed and employed in order to simulate reflection and scattering of FM signal from the

disk surface and edges. By rotating the disk around its central axis reflected echo patterns from its edges

change allowing for small time variation between arrivals of each individual echo component. This repre-

sents the scenario of a flying insect where the distances from the bat to the insect’s head, body and wings

are slightly different with each returning a contribution to the overall echo. The echolocation signal obtained

from the rotating disk is further encoded into the spectrogram-like format characteristic for the mammalian

auditory system. The simulation results presented in this paper demonstrate that the proposed model is able

to distinguish between overlapping echoes from the spectrogram-like forms of the echolocation signals.

1 INTRODUCTION

Most bat species use echolocation to navigate in to-

tal darkness and/or to track and capture flying insect

prey by processing acoustic information based on

the reflection of sound waves from targets and inter-

cepting obstacles. These mammals can perceive the

distance to a target based on the delay between the

emitted pulse and the returning echoes with surpri-

singly fine delay acuity and are also able to deter-

mine the target position from the different arrival ti-

mes, intensity levels and spectral information of the

returning echoes. Furthermore, some species of bats,

such as the big brown bat Eptesicus fuscus, emit

frequency-modulated (FM) ultrasound of very short

duration which they use as broadband signals to

reconstruct images of the arrangement of reflecting

points and surfaces within the target. The complex

structure of the echo formed by sound scattering

from the discrete target points contains interference

peaks and notches at specific frequencies determined

by the time separation of the individual reflections.

This overall echo pattern is altered further due to the

relationship between the bat and target movements.

The phenomena that echolocating bats have

evolved, both behavioural and physiological mecha-

nisms, to resolve those difficult tasks have inspired

researchers to thoroughly study their behaviour and

understand the underlying physical principles and

signal processing techniques with the aim of develo-

ping a computational model of the echolocation pro-

cess which can be adopted in sonar system design in

the future.

The objective of this study is to develop a com-

putational echolocation model that can be used for

discrimination of the specific target object and re-

construction of the acoustic information from multi-

ple overlapping echoes returning back from that tar-

get. An acoustic model for estimation of the back-

scattering impulse responses of a rigid disk has been

developed and employed in order to simulate reflec-

tion and scattering of FM ultrasound signal from the

disk surface and edges. By rotating the disk around

its central axis reflected echo patterns from its edges

change allowing for small time variation between

arrivals of each individual echo component. This

represents the scenario of a flying insect where the

distances from the bat to the insect’s head, body and

wings are slightly different with each returning a

contribution to the overall echo. The resulting signal

containing both emission and echoes from the disk is

further encoded in the spectrogram-like format using

a set of broadband filters adapted to the properties of

the mammalian auditory system (Sailant et al, 1983).

300

Nikoli

c D., Papadopoulos T. and Allen R. (2010).

USING BAT AUDITORY MODELLING FOR OBJECT DISCRIMINATION AND ECHO SEPARATION TASKS.

In Proceedings of the Third International Conference on Bio-inspired Systems and Signal Processing, pages 300-305

DOI: 10.5220/0002748603000305

 SciTePress

This paper is organised as follows: A detailed

description of the proposed model is given in section

2. Simulation results obtained at different processing

stages explained in section 2 are presented and dis-

cussed in section 3. In section 4, the obtained results

are summarised and some directions for the future

work are outlined.

2 MODEL DESCRIPTION

The proposed simulation model shown in Figure 1

consists of three processing blocks: the system mo-

del, the bioinspired cochlear block and the artificial

post-processing block. The important stages of each

block are described in the subsequent sections.

Figure 1: Simulation echolocation model used for object

discrimination and echo separation.

2.1 System Model

The system model is used to simulate echoes which

bats might receive from a specific disk-shaped

target. This is achieved by separately modelling the

emission, backscattering and reception of the ultra-

sound signal in the field. Emission is modelled using

a hyperbolic, linear period-modulated down-swept

signal in the range of frequencies from 100 kHz to

20 kHz. Time duration of this signal used for all

simulation sets was 1 ms with 300 μs rise-fall time

and unity signal amplitude. The sampling frequency

was set to 100 MHz. This signal is a close replica of

the FM signal commonly used by some bat species

(FM signal emitted by the big brown bat Eptesicus

fuscus is within the same frequency range).

The input signal is further filtered through the

synthetically-generated backscattering impulse res-

ponse of the rotating disk, to obtain reflection com-

ponents associated to a relative position of the disk

to the source and the receiver. The output of the sys-

tem model block consists of the emission and echoes

produced by the disk positioned at a certain angle re-

lative to the axis perpendicular to the source-receiver

axis. The effects involved in the reception of the sig-

nal such as those from the bat’s head and ears are

being considered in a parallel study.

2.1.1 Analytical Modelling of the Disk

Backscattering Impulse Response

The physical setup considered here is that of an ideal

point monopole source, an acoustically rigid disk-

shaped target of finite depth D between its two cir-

cular surfaces of radius r and a receiver positioned

along the line joining the source and the center of

mass of the disk (Figure 2). The disk is positioned

such that the axis parallel to its circular surfaces and

passing through its center of mass is perpendicular

to the source-receiver axis. Cases are considered

where the disk target is rotated around the axis that

is parallel to its circular surfaces.

Figure 2: System model geometry used for estimation of

the backscattering impulse responses of the disk target.

The emission-backscattering-reception process is

modelled as the output of a linear time-invariant sys-

tem. The input to this system is the source strength

q(t), taken here to be equal to the chosen FM signal.

The system’s output, i.e. the acoustic pressure p(t)

created at a single point of the generated sound field,

is modelled by the convolution integral:

() ( )( )

() () () ( )

∞

=−

⎡⎤

=++ −

⎣⎦

∫

dir ref diff

pt h qt d

hhhqtd

τττ

ττττ

(1)

where h

dir

and h

ref

describe the direct propagation

from the source to the sampling point and the specu-

lar reflection from the planar disk surface. They

amount to delayed and attenuated Dirac delta pulses

= δ(τ–d

/c)/d

with d

denoting the corresponding

path length distances and c the speed of sound. The

performance of bioinspired time-frequency auditory

models using such geometrical-acoustics-derived in-

put signals has been examined in numerous studies

in the past (see (Neretti et al, 2003), and (Saillant et

al, 1993) and references therein). A novel approach

in this study is the inclusion of the third term h

diff

which models the contribution of diffraction from

USING BAT AUDITORY MODELLING FOR OBJECT DISCRIMINATION AND ECHO SEPARATION TASKS

301

the circular edges of the target. It should be clear

that as the disk is rotated from the source-receiver

axisymmetric position, the point on its circular face

that contributes to specular reflection progressively

moves towards the circumference and eventually

meets it at certain angle of rotation. For larger rotati-

on angles, the time-concentrated part of impulse res-

ponse due to h

ref

vanishes, leaving only time-extend-

ed contribution of h

diff

in the generated echo. The be-

haviour of the proposed model for such time-smear-

ed echoes is one of the objectives of the present

study.

For the computation of the diffraction part of the

impulse responses the time-domain solution present-

ed in (Svensson et al, 1999, 2006) was used. This is

based on the use of directional secondary sources

positioned along a finite-length edge, with strength,

timing and directivity adjusted to conform to pre-

vious solutions of the infinite-length edge diffraction

problem. The diffraction impulse response h

diff

thus derived as a line-integral over the length of the

edge of the contribution of those secondary sources.

A set of Matlab scripts for the determination of the

edge parts that contribute to the overall diffraction

impulse response, the derivation of the related

directivity functions and timings and the discretisa-

tion of the line integral computation provided in

http://www.iet.ntnu.no/~svensson/downloads/ was

used for the computation of the impulse responses in

this study. The circular faces of the target were

approximated with a 256-sides polygons and the

sampling rate was set to 10 MHz. Only first-order

diffraction was included in these results. A sampling

rate of 100 times the highest frequency of interest

(100 kHz) was chosen in order to minimise the

effect of the discretisation of the line integral com-

putation (Svensson et al, 1999).

2.2 Bio-inspired Cochlear Block

The auditory processing of the echo signal reaching

the bat’s ear after returning from the target object is

performed by the cochlear block, named after the

main auditory part of the inner ear known as the

cochlea. A major function of the cochlear block is to

model the sound signal that arrives at the outer ear

canal into the spectral format that is characteristic

for the mammalian auditory system. The frequency

selectivity and tonotopic organisation of the basilar

membrane has been modelled by the gammatone

filterbank (Patterson et al, 1992). The shape of the

auditory filters is characterised in terms of the equi-

valent rectangular bandwidth (ERB) scale (Glasberg

and Moore, 1990). This has been widely used appro-

ach in similar modelling work to represent auditory

filters in the peripheral auditory system of the mam-

mals and humans.

The gammatone filterbank has been designed as

a series of 81 eight-order IIR gammatone bandpass

filters with center frequencies spaced from 20 kHz

to 100 kHz. The implementation of this filterbank is

based on (Slaney, 1993) and modified to accommo-

date the frequency range of interest for this study.

Although there is no agreement regarding the den-

sity of frequency channels in the auditory filter and a

varying number of filters is employed in different

frequency ranges, the number of filters that has been

chosen in this study appears to be reasonable compa-

red to similar models (Sailant et al, 1983).

The mechanical motion in the basilar membrane

that resolves the frequency is further converted into

neural activity by the inner hair cells. This stage is

modelled by half-wave rectification followed by the

1st-order low-pass filtering with the cut-off frequen-

cy of 3 kHz applied in order to remove unwanted

frequency components generated by half-wave rec-

tification.

2.3 Artificial Post-processing Block

The concept of the bioinspired cochlear block results

in the multichannel output signal that represents the

auditory spectrogram-like image of the FM transmi-

ssion and echoes. It is known that the bat has ability

to transform such perceived acoustic image from the

spectral domain into spatial image that explicitly re-

veals the placements of echoes along the range axis

showing the small differences in the distance to dif-

ferent parts of the target (Saillant et al, 1993). Since

echo delays are related to distances to different parts

of the target object, by acquiring these delays infor-

mation related to the object range, size and shape is

also obtained.

In order to extract information about delay of dif-

ferent components of complex FM echoes from an

auditory image derived from the cochlear block in

the present study, it is necessary to additionally pro-

cess the output of the cochlear block. Thus, the pur-

pose of the further processing in the artificial block

is to obtain an image constructed from the delays of

different components of complex echoes.

3 RESULTS AND DISCUSSION

The backscattering impulse responses of the four

circular disks with the radiuses set to 30, 50 and

150 mm and thicknesses of 1, 15, 30 and 50 mm

BIOSIGNALS 2010 - International Conference on Bio-inspired Systems and Signal Processing

302

were calculated based on acoustic modelling descri-

bed in section 2.1.1, and used for further simulation

purposes. The relative distances between the source,

receiver and the disk center of mass, are kept

constant through all simulations. The main geo-

metrical parameters used for simulations are listed in

Table 1.

Table 1: The parameter sets used in simulations.

Set no. r [mm] D [mm] d

[cm] d

[cm]

1 30 1 10 150

2 50 15 10 150

3 150 30 10 150

4 150 50 10 150

For each given set of parameters, the disk place-

ment relative to the source-receiver axis is changed

by rotating the disk around its lateral axis which

results in different impulse response. The rotation

angle α is varied from 0 to 90° with 1 degree step.

The value of α=0° corresponds to the position where

the circular surfaces of the disk are perpendicular to

the source-receiver axis (marked as shaded region in

Figure 2) and α=90° corresponds to the position

where the circular surfaces are parallel with the

source-receiver axis.

In Figure 3, the estimated backscattering impulse

responses are plotted for the parameters nominated

as set 3 in Table 1 but stacked with a vertical offset

of 6x10

-5

and zoomed-in in the time interval of the

backscattering part. These responses are low-pass

filtered through a first-order Butterworth filter with

the cut-off frequency set to 200 kHz. This figure

shows the evolution of the h

ref

and h

diff

parts of

Eq. (1) as the disk rotation angle increases. As

mentioned above, the component of the impulse res-

ponse h

ref

becomes negligible and diminishes as α

increases leaving only the contribution of h

diff

in the

reflected part of the impulse response. For α=0°

(top-most response in Figure 3), the impulse respon-

se comprises a positive delta pulse due to the spe-

cular reflection from the circular face of the disk

followed by a negative delta pulse due to the

contribution of the edge diffraction from the circular

edge of the disk. In the α=0° axisymmetric geome-

try, this edge diffraction contribution is time-aligned

and hence appears concentrated in a single pulse. As

soon as the disk is rotated from the axisymmetric

degree, this time-alignment of the diffraction con-

tribution is disrupted, and the negative pulse

progressively splits into two parts, their distance

being determined by the path length difference bet-

ween the source to the leading or trailing edge-point

of the disk and back to the receiver. Furthermore, the

specular reflection pulse moves forward in time and

becomes attenuated as the reflection point on the

circular face of the disk moves outwards from the

center. For a rotation angle of α=6°, the specular

reflection part h

ref

vanishes leaving only the diffra-

ction part h

diff

in the plotted impulse responses. This

part evolves in three parts with increasing angles of

rotation. The onset of these three parts correspond to

the geometric paths from the source to the front

leading (FL) edge, back leading (BL) edge and front

trailing (FT) edge of the disk (Figure 2).

    





 



 







D 

Figure 3: The backscattering impulse responses calculated

for the parameters designated as set 3 in Table 1.

The objective of the results and analysis present-

ed in the following part of the paper is to identify

these features in the spectrogram-type output of the

cochlear block output. The generated down-swept

FM signal is filtered through the backscattering

impulse response estimated for the particular disk

geometry. The impulse response of the direct

transmission path between the source and receiver

(delta pulse) is normalised to unity while the back-

scattering part of the impulse response is normalised

to 0.25. This is done with intention to enhance the

relatively small contribution of the backscattered

part to the overall impulse response and does not

have a negative affect on the results since the main

goal is to distinguish between the reflected echoes

based on their arrival times. Moreover, this approach

imitates what bats do by contracting the middle ear

muscle to attenuate the sensitivity of the ear and

decrease the intensity of the emitted pulses to the

certain level to protect the hearing system from

overstimulation (Kick and Simmons, 1984). The

resulting echolocation signal encompassing both

emission and the complex echo signal is further pro-

cessed through the gammatone filterbank and de-

composed into 81 filter channels. Each gammatone

USING BAT AUDITORY MODELLING FOR OBJECT DISCRIMINATION AND ECHO SEPARATION TASKS

303

filter is followed by half-wave rectifier and first-

order low-pass filter with cut-off frequency of 3 kHz

producing smoothed envelopes across all channels.

In order to classify the echo components present

in the spectrogram, the output of the cochlear block

is dechirped and the integration over all frequency

channels aligned in time is performed. This proce-

dure describes the function of the artificial block.

The local maxima in the overall smoothed envelope

correspond to main echo components present in the

backscattering part of the echolocation signal

(Figure 4).









 

     







 

 

Figure 4: The received echo signal (top), the dechirped

output of the cochlear block (middle) and the output of the

artificial post-processing block (bottom) obtained for the

parameters designated as set 3 in Table 1 and the rotation

angle α=30°.

The overall signal envelopes calculated for each

simulation set and the angle of rotation are plotted in

Figure 5. Two echo components from the leading

and trailing disk edges are evident in all simulation

sets for rotation angles larger than 10 degrees. The

distance between these two echoes corresponds to

the disk radius. For the disk of 50 mm radius and

15 mm thickness (set 2), the main echo component

formed by the leading disk edge splits into two sub-

components as a result of diffractions from the front

and back face edges (Figure 5a). This effect can not

been seen for the disk of only 1 mm thickness (set 1)

since the time delay between those two echo compo-

nents is less than 5 μs for all rotation angles (Figure

5b). However, for the sets 3 and 4 (Figure 5c-d),

where two disks of larger size are used, the separa-

tion effect between echo components formed by the

front and back leading edges is more significant.

Finally, time delays between the separated

echoes obtained using the model proposed in this

paper are compared with the real geometry of the

disks used in this simulation. The results of this

assessment are presented in Figure 6. Here, the time

differences between the front leading and front

trailing edges (FT-FL) and the front leading and

back leading edges (FL-BL) are calculated from the

disk geometry for each angle of rotation and compa-

red with the time delays of the main echo compo-

nents, denoted as Δt

and Δt

, derived from the

model outputs plotted in Figure 5. For this illustra-

tion, peripheral points on the front and back disk

circumferences which are farest from the axis of

rotation are chosen to represent each of these edges.

Therefore, the time differences between them are

calculated based on the path length distances from

the source to the each of these points and back to the

receiver. The rotation angle for which the back lead-

ing edge of the disk become visible is denoted as α

The main echo components are defined as local ma-

xima in the overall signal envelope. A close match-

ing between the results obtained by using the audi-

tory images with the ones defined by real geometry

of the disk is evident for all 4 simulation sets. These

demonstrate that the auditory images directly repro-

duce the geometrical parameters of the disks since

the time delays Δt

and Δt

are linked to the disk

radius and thickness.

4 CONCLUSIONS

The bat-inspired echolocation model developed in

this study and presented in this paper demonstrates

ability to separate multiple overlapping echoes that

are present in the echolocation signals formed by the

rotating disk-shaped target object. A close matching

between the results obtained by using this simulation

model with the ones defined by real geometry of the

disk proves that the proposed model has ability to

distinguish between overlapping echo components

from the disk edges and, therefore, to discriminate

between different disk-shape targets.

Further study has been carried out to enhance

the performance of the proposed echolocating model

by employing more advanced signal processing

techniques in the analysis of the spectrogram-like

auditory images. It would be also of interest to test

this model using experimentally measured back-

scattering impulse responses of different targets at

different orientations.

BIOSIGNALS 2010 - International Conference on Bio-inspired Systems and Signal Processing

304

(a) Set 1 (b) Set 2

Figure 5: Auditory images of the echolocation signals

formed by the rotating disks defined in Table 1.

         







D 

  P



  



'



'



         





D 

  P



  



'



'



(a) Set 1 (b) Set 2

         







D 

  P



  



'



'



         







D 

  P



  



'



'



Figure 6: Time differences between the front leading and

front trailing edges (FT-FL) and the front leading and back

leading edges (FL-BL) of the disk and the time delays (Δt

and Δt

) derived from the auditory images shown in

Figure 5.

ACKNOWLEDGEMENTS

The authors are very grateful to RCUK for support

through the BIAS Basic Technology Programme.

We gratefully acknowledge the use of the EDTB

code made publicly available by Prof. U.P. Svensson

of the Norwegian University of Science and Techno-

logy. The authors would also like to thank R. Collier

for proofreading of this paper and useful comments.

REFERENCES

Glasberg, B. R., and Moore, B. C. J., 1990. Derivation of

auditory filter shapes from notched noise data. Hear-

ing Research, 47(1-2), pp. 103–138.

Kick, S. A. and Simmons, J. A., 1984. Automatic gain

control in the bat’s sonar receiver and the neuroethol-

ogy of echolocation. J. Neurosci. 4, pp. 2725-2737.

Neretti, N., Sanderson, M. I., Intrator, N., and Simmons,

J.A., 2003. Time-frequency model for echo-delay

resolution in wideband biosonar. Journal of Acoustic

Society of America, 113(4), pp. 2137-2145.

Papadopoulos, T. and Allen, R., 2007. Experimental

method for the acoustical modelling of the echoloca-

tion process in bats. Proceedings of the Institute of

Acoustics, 29(3)

Patterson, R. D., Robinson, K., Holdsworth, J., Mckeown,

D., Zhang, C., and Allerhand, M. H., 1992. Complex

sounds and auditory images, Auditory Physiology and

Perception, (Eds.) Cazals, Y., Demany, L., Horner. K.,

Pergamon, Oxford.

Saillant, P. A., Simmons, J. A., and Dear, S. P., 1993. A

Computational Model of Echo Processing and Acous-

tic Imaging in Frequency-Modulated Echolocating

Bats – the Spectrogram Correlation and Transforma-

tion Receiver. Journal of Acoustic Society of America,

94(5), pp. 2691-2712.

Simmons, J. A., Saillant, P.A., Wotton, J.M., Haresign, T.,

Ferragamo, M.J., Moss, C.F., 1995. Composition of

biosonar images for target recognition by echolocating

bats. Neural Networks, 8(7-8), pp. 1239-1261.

Slaney, M., 1993. An Efficient Implementation of the

Patterson-Holdsworth Auditory Filter Bank, Apple

Computer Technical Report 35, Perception Group-

Advanced Technology Group (1993).

Svensson, U.P., Fred, R.I., and Vanderkooy, J., 1999. An

analytic secondary source model of edge diffraction

impulse responses. Journal of Acoustic Society of

America, 106(5), pp. 2331-2344.

Svensson, U.P. and Calamia, P.T., 2006. Edge-diffraction

impulse responses near specular-zone and shadow-

zone boundaries. Acta Acustica United with Acustica,

92(4), pp. 501-512.

USING BAT AUDITORY MODELLING FOR OBJECT DISCRIMINATION AND ECHO SEPARATION TASKS

305