Distributed Mode Loudspeaker Polar Patterns 5065 (E-6)

James A. S. Angus
University of York
York YO1 5DD, UK

Presented at ^uD,o
the 107th Convention
1999 September 24-27
New York
Thispreprinthas been reproducedfrom the author'sadvance
manuscript,withoutediting,correctionsor considerationby the
Review Board.The AES takesno responsibilityfor the
contents.

Additionalpreprintsmay be obtainedby sendingrequestand

remittanceto the AudioEngineeringSociety, 60 East42nd St.,
New York,New York 10165-2520, USA.

All rightsreserved. Reproductionof thispreprint,or anyportion

thereof,is not permittedwithoutdirectpermissionfrom the
Journal ofthe AudioEngineeringSociety.

AN AUDIO ENGINEERING SOCIETY PREPRINT

 Distributed Mode Loudspeaker Polar
Patterns
JAMES A S ANGUS

Department of Electronics, University of York,York, England.

The Distributed Mode Loudspeaker's polar response is analysed with

reference to it's radiation mechanism. In particular the polar response
below and above the supersonic plate propagation boundary is examined.
The paper shows that a quasi omnidirectional response is possible.

0 INTRODUCTION
The Distributed Mode Loudspeaker (DML) has been touted as a near perfect
omnidireectional source. This is not strictly speaking true. The purpose of this
paper is to provide an assessment of the polar response of a Distributed Mode
loudspeaker by considering its radiation mechanism. In particular the paper will
analyse the difference in polar pattern that occurs between the plate wave being
subsonic as opposed to supersonic. Simulated result for these situations will be
presented and the paper will show that quasi omnidirectional performance is
possible, but only under certain conditions. The paper will first briefly examine
radiation mechanisms and then look at the polar patterns which arise. Finally it
will examine published polar pattern data.

I RADIATION FROM PANELS

z lI air I_ 2= t_J
I I k_,_.,e I
I /-%'
· %/ C plate -- b
o
%/=
'_pltlle

Figure 1 Sound radiation from waves on an infinite plate.

The basic principle of radiation from a plate or membrane is illustrated in

figure 1 which shows a travelling wave in an infinite membrane in contact with
air. In order for sound energy to be radiated there must be a lateral
displacement of the air. Clearly the air and the plate variation in the
longitudinal (or x) direction must be equal because they are in contact.
However the lateral displacement of the air will depend on the velocity of sound
in the air compared to the phase velocity of the wave in the plate. In simple
terms the phase velocity in the must be faster than the speed of sound in the air
or else the air will be able to move out of the way as the wave passes and
therefore not propagate a sound wave. The best way analyse this situation in
more detail is to look at it using wavenumbers. By using wave numbers one can
show that the wave number in the transverse (or z) direction, which corresponds
to the wave propagating ii1 the air, is given by:

This equation has three possible conditions:

kpl_,_< kai,. Here the wave causes sound radiation from the plate, a smaller k
implies a higher phase velocity at a given frequency. These sound
waves propagate away from the plate so the sign in equation 16 is
positive. The wave propagates outwards at an angle 0 from the
normal given by:

ky kplate (2)
cos_ - ki, -
Note that this is a very directional wave propagating in one
direction. This means that, contrary to what is publicly stated, it is
possible for distributed mode loudspeakers to be directional. It is
also interesting to note that the if the panel wave speed is infinite
then the direction of propagation is normal to the plate. One can
view this condition as the condition required for piston radiation.

The wave impedance fro this propagating wave at the plate surface,
remembering that the plate is infinite, is given by:

(°P° P°C (3)

- kplat
e kplate

Zwave _a2ir 2 _ ( k_/r )2

This impedance is positive and real and therefore work is done on

the air, energy transferred and therefor sound energy propagated.

kpt_te > k_,i,. Here the wave speed in the plate is less than the speed of sound in
the air. This means that the result of equation 1 is imaginary and
no wave propagation results: that is:
 . 2 2
kz = -j _[kp2late - kalr (4)

This means that the disturbance of the air decays exponentially

away from the plate surface. This is another way of saying that the
air can "get out of the way" when the speed of the wave in the plate
is less than the speed of sound in air.
The wave impedance at the surface is given by:

jp°c (5)
-
_(kair j 1

This impedance is purely reactive and so no work is done, however

it can affect the vibration of the plate via inertial loading if the
plate is light enough.

k_lat
e =kai r In this condition the two velocities are equal however this
condition cannot be satisfied physically because the wave
impedance at the surface is infinite.

Thus the implication of these equations are that unless the phase velocity of
the wave in the plate is greater than the speed of sound in air it will not radiate'
efficiently. Furthermore when the plate is radiating efficiently the radiated
sound is directional. This presents us with a problem because the density of
modes in the plate depends on the speed of sound in the plate. If the wave
velocity in the plate must be greater than the speed of sound this will set a lower
resonance frequency which will be invariably greater than the lowest audio
frequency for reasonably sized plates. In order to have a mode at 34 Hz while
still remaining above the critical frequency requires a plate dimension of greater
than 10m! Note that this requirement is the same whether bending or transverse
shear waves are involved. Clearly this is not the whole story because plates do
radiate when the plate velocity is less than the speed of sound in air. The reason
for this is that the above analysis assumes a plate of infinite extent. However if
the plate is of a finite size there are effects due to the boundaries which are
explained in the next section.

!_ aa--
m

I [

[ I

Figure 2 A finitewidth radiatingstripmode.

1.1 Sound radiation from a finite width strip.
I ! I

Resulting
waveform ._
7FHz
5 FHz
3.33FHz
3 FHz
2.66FHz

1,33FHz

I I I

0.0 0.5 1,0 1,5 2.0

Time (in cycles)

Figure 3 Construction of a pulse from sine wave components.

In an infinite panel waves propagate in one direction forever and exist over an
infinite area. However in a finite panel the waves only exist over a finite area
and have zero amplitude elsewhere. Figure 2 shows this for the one
dimensional case of an infinite strip in an infinite baffle. We can analyse this
situation as a superposition of forward and backward travelling waves and treat
them separately. However we need to have some way of expressing the fact
that the waves are of zero amplitude outside the plate. One way of doing this is
to analyse it from a Fourier point of view. In the time domain we can construct
a pulse from an infinite summation of sine waves, as shown in figure 3 In the
spatial, or plate domain, we can also construct a finite width propagating wave
from an infinite number of propagating waves of the same frequency but
different phase velocities. These form components with different values of k
the summation of these components is known as a wavenumber transform.
Mathematically we can say that the waves on the plate are given by:

f(x)= _I_2 F(k)exp(jkx)dk (6)

F(k) .[_ f (x) exp(-jkx)dx (7)

The negative values for k correspond to waves propagating in the other

direction.

4
 /
"" ,,,
_,
J avefront
direction
/ Wavelength X // (oft.angential
/ (due to plate / mode)

ve.phaSev_ wa. /
Phase wavelength ;_
(due to apparently
higher phase velocity)

Figure 4 The geometry behind higher effective phase speeds in finite strips.

But how can one have an infinite set of phase velocities physically? One way
of looking at this problem is shown in figure 4 which shows a wave propagating
along an infinite strip. Because of the finite boundaries there are an infinite
number of possible ways for the wave to propagate between the strips from
normal to parallel to the boundaries. As shown in figure 4 a wave which
propagates at an angle between has a wavelength which corresponds to the
phase wavelength in the direction of propagation. However due to the geometry
of the propagation the effective wavelength along the boundary is greater and is
given by:

_plate (8)
/_effective COS 0

Although the wavelength is longer the frequency is the same and therefore
the effective phase velocity will be greater and thus k will be smaller. For the
strip geometry the radiation direction will be normal to the boundaries and
therefore it will be the k values in the x direction which determine the radiation
efficiency and these can also be both smaller and greater than that purely due to
the medium's phase velocity.
 a

I I

Figure 5 The mode shape for an mth order mode.

Consider the mode shape in the x direction shown in figure 5 for an mth
order mode. The normal velocity for a strip is give by:

v(x't) = {; sin(-m-ff-_)exp(jc°t) 0>x>a0-<x---a (9)

and the corresponding wavenumber transform is given by:

V(kx) = vsin exp(-jkxx)dx (10)

This is equivalent to the Fourier transform of a rectangularly weighted sine

pulse which is a sinc function centred around the frequency of the sine wave.
As we are interested in the radiated energy we need to examine the squared
modulus of this wavenumber transform and this is given by:

IV(kx>12--Ivl2 }/
I kx2 2_k a j j]2sin2(kxa2mlr ¢

Equation 11 is called a modulus-squared spectrum and is proportional to

energy. This representation is highly useful for examining the radiation
efficiency from a finite width strip. Recall that the condition for radiation is that
the phase velocity of the waves must be greater than the speed of sound in air,
that is kx < mair.For a given mode frequency Om the value of _+k.i,.
(_kal r =q-Om/Cai,. will be a particular value on the kx axis of the modulus-
)
squared spectrum. Values of kx which are less than this will contribute to
radiation whereas values of k_ above this value represent reactive energy which
creates nearfield disturbances near the plate but does not contribute to radiation.
Thus one can mark out a region of the modulus-squared spectrum which

6
contributes to this radiation that is the total radiated energy will correspond to
the total area under the modulus-squared spectrum from -k,i ,. to +k_i,. and the
radiation efficiency will be the ratio of this area to the total area under the
modulus-squared spectrum, as shown in figure 6. The modulus-squared
spectrum can also be used to examine the effect of varying the plate parameters
on there radiated power. Figure 7 shows the effect of increasing the phase
velocity of the wave in the plate while retaining the same mode order and shape.
As this would correspond to a higher frequency the value of _+k_,i,. will also
increase, as shown in figure 7 and thus the radiated energy will also increase. If
the phase velocity of the plate wave increases enough the main lobe of the
modulus-squared spectrum will form part of the radiated power. At this point
there will be a rapid increase in the radiated power as shown in figure 8 which
shows the area under the modulus-squared spectrum from -k x to +kx as a
function of kx. This condition correspond to a supersonic phase velocity for
plate wave propagation. Clearly if this condition occurs over only part of the
frequency range a nonuniform radiated frequency response will result. Figure 9
show the effect of a having different mode order, and therefore different
frequency, in a strip of constant width and plate phase velocity. The important
thing to note is that although the value of +k,i,. increases the centre of the main
lobe also increases in proportion. Figure 10 shows the effect of a wider width
of the strip while retaining the same plate wave phase velocity. From these
figures we see that the total radiated energy changes slowly. This is a surprising
result that is not immediately obvious. However it can be explained by
examining figures 11 and 12 which show that the part of the plate which
contributes to the radiation is the quarter cycle at the strip edges. This is
because for subsonic plate velocities the cycles within the plate cancel each
other out. Because of this the effect of increasing the width of the strip is small.
However, note that as the strip gets wider the width of the main lobe and
sidelobes get narrower and in the limit the main lobe and sidelobes become a
Dirac delta at the +_kpt,t e value corresponding to the modal frequency when the
strip width becomes infinite. Under these conditions the radiation efficiency is
zero if --+kalr is less than +kp_at _ which corresponds to the results obtained earlier
for an infinite plate.
This section has shown that a finite width plate can radiate sound even if the
plate phase velocity is less than the speed of sound in air, although the
efficiency is considerably reduced over that for supersonic plate wave
velocities.
 Figure 6 The radiating wavenumber spectrum components of a mode.

w,

Figure 7 The radiating wavenumber spectrum components for the same mode at
a higher frequency and hence higher plate velocity.
 vv

Figure 8 The integrated wavenumber spectrum of a mode.

Figure 9 The radiating wavenumber spectrum components of a higher order

mode.
 ¥

Figure 10 The radiating wavenumber spectrum components of the same mode

with a wider strip.

Figure 12 The effective radiating components of an odd subsonic mode.

10
1.2 Sound radiation from finite homogeneous plates.
The concept of the wavenumber spectrum can be extended to finite plates as
follows. For a rectangular plate of dimensions a x b the normal velocity for a
mode is given by:

v(x,y) =
vcos cos {°o<Xa
< y< b
(12)

0 >y>b
{00>x>a

vsin sin
v(x,y) = 0 < y< b (13)

0 0>y>b
0>x>a

The corresponding wavenumber transform will be two-dimensional and is

given by:

V(kx,ky) = vsm (14)

ii
00 . (m__.).sm (n.__)exp(-jk,,x)exp(-jkyy)dxdy
. .

For a rectangular plate this transform has a similar result to that of a strip, that
is a sine squared shape centred at wavenumber values corresponding to the
modal frequencies in both the x and the y directions. The main difference being
that the sine squared function is now two dimensional and the centre of the main
lobe is now at the intersection of the kx and ky values for the mode. As in the
strip case values of k(x,y) which are less than the value of +kalr will contribute
to radiation. In the two dimensional case this condition describes a circle of
radius k_,i,,around the origin. Unlike the one dimensional strip it is possible for
one dimension to have a supersonic wave velocity while the other does not. In
addition both dimensions may be subsonic or supersonic. In fact any condition
which satisfies the following condition is possible:

kx
2 + k2 < k[,2ir (15)

These conditions are shown in figures 39, 40, 42 and 43 as contour plots of
one quadrant of the two dimensional wavenumber transform. As this transform
has fourfold symmetry so this is all that is required for display. A circle
quadrant marked on each graph represents k,,r and any components which lie
within this quadrant will contribute to radiation.

11
Figure 13 shows the case of the phase speeds in both directions being subsonic.
This condition corresponds to corner radiation because it is the corners of the
plate which contribute to radiation. This configuration is the two dimensional
equivalent of the edge radiation from the strip and we would expect similar
behaviour from it. That is if the mode frequency is higher for a given mode
shape then the radiation efficiency is increased. The radiation efficiency will
also stay fairly constant as the mode order, or the plate size, is increased.

. +
- + - +
-I- - -1- -
- -t- - -I-
-i- _ -t- .

- -I- - -l-

-t- . -i- .

,xX%" + -
Figure 13 Typical plate vibration pattern for a corner mode.

,xXXXXX_ ,x'x'x'x'xXX'%'x'x'xX'x'xX_
x
- +
+
- +
+
- +
+
_'x'xNX_x'xXX-%,X'xNNX'_
,x'x,'xxxX
'xXX'_,X_
Figure 14 Typical plate vibration mttern for an x direction edge mode.

12
 Figure 15 Typical plate vibration pattern for a y direction edge mode.

Figures 14 and 15 show the cases in which one of the edge directions has a
supersonic phase velocity, remember this is a function of geometry. In these
cases the relevant edges contribute to the radiation. This mode of radiation is
more efficient than the corner radiation mode described previously. In all these
subsonic radiation situations it is the edges that provide volume velocity to
contribute to radiation. Unlike the corner modes increasing the plate dimension
in the strip direction will increase the radiating efficiency because there will be
more area at the edges to contribute to the radiated energy.
These results apply to single mode excitation in a panel. However in practice
one would be exciting several modes at once and this makes it harder to see the
radiation efficiency as a function of frequency. In order to overcome this
problem an average over a small frequency band, called a modal average
radiation efficiency. To do this an assumption has to made regarding the
distribution of the excitation energy among the modes and the most common
assumption is that of equipartition of energy among the modes.

1.3 Sound radiation from finite sandwich plates.

For non homogeneous materials, in particular those consisting of two thin
plates separated by a core of lower shear stiffness, for example Aerolam, the
radiation mechanisms are the same. However the variation of plate phase speed
with frequencies is different and this does affect the frequency response of the
radiation. This is because there is no longer a single propagation mechanism for
the plate wave. At Iow frequencies the plate wave is a bending wave whose
velocity is determined by the bending stiffness of the whole panel. Because the
panel will be quite thick, 12mm or more, and because the core has a relatively
low shear modulus, the bending wave will become a shear wave whose velocity
is determined by the core shear modulus. This will happen at intermediate
frequencies. At higher frequencies the bending wave propagation again occurs
due to the individual faceplate bending stiffness. The net result of all this is that

13
the combination of these mechanisms results in a speed of sound which is less
than the speed of sound in air for a wider frequency range than would be the
case if the panel was homogeneous. This means that, by appropriate choice of
material properties, the plate can be operated just below critical frequency for a
wider frequency range. It also means that practical panel speakers will be
operating with a subsonic wave speed over most of their frequency range except
at the higher frequencies.

2 THE POLAR PATTERN OF THE RADIATION FROM A PLATE.

The polar pattern of a distributed mode radiator depends on the mode of
radiation. The subsonic radiation modes have a different polar pattern
compared to supersonic modes. At low frequencies only the edges are
radiating, and are separated by several wavelengths. Therefore the polar pattern
will be the same as that of spaced sources, that is, an interference pattern, as
shown in figure 16 for an odd mode in a strip radiator. As there are two
possible edge radiation configurations there are two possible polar patterns and
the second, even mode, one is shown in figure 17. These two patterns are
orthogonal and as the frequency varies the radiation pattern will swap from one
to another depending on the mode structure. Figure 18 overlays the two
patterns to show their differences. For a plate radiator operating as a corner
mode radiator one would get a similar effect in two dimensions and so the polar
pattern would resemble a twitching hedgehog. Superficially this gives a quasi
omnidirectional response but in reality some strong frequency variations are
occurring at a given point in space. As ones ears are at two different points in
space one may hear a uniform response because they average out the variation
but the accuracy of stereo imaging is likely to be affected.

Figure 16 The polar pattern of an odd order edge mode in a strip radiator

14
 r_
L1 Jl
I· IIZ & J#r I JlA
1· L I J ! I!
·Ll L II I: II JJ
rJ .
I· L · 1 f r · I ti
· x L Z I I r I _ J _'r
L· z · t _ z ! i zJ

· _ · · L · L · I J J W J · · _#
_ _ · · _ · I I I J I r · · · s #

Figure 17 The polar pattern of an even order edge mode in a strip radiator

i_ IL JI JI
I1 I I I I sr .
Iq II I1 J I I I el

. t % i i i i i i ! ! f J .
ITL I 'tL % T I I I I r · · ! fl

r _ · _ tJ
%-. ',% ,,- , J i · r xr
%%. ,,._ , r! / //

d ', ,, .

i-.' ? ' J'

\
Figure 18 The polar pattern of odr and even order edge modes overlaid.

For supersonic modes the radiation pattern is very different. In this case all
the surface contributes to radiation and the forward and backward waves radiate
in their direction of propagation at an angle from the normal given by the
equation:

0 = cos-I 1 -- kair

15
 Equation 16 shows that as the plate phase speed increases, kpla_e gets closer to
zero, the angle of radiation approaches the normal direction. The effective
radiation area is a rectangular aperture and so the radiation pattern would be a
sinc(x) type pattern and in a strip radiator there would be two of them pointing
in opposite directions, corresponding to the forward and backward waves. This
is shown in figure 19. Figure 20 shows the effect of doubling the strip width
and one can see that the polar pattern is considerably narrowed. One would get
a similar effect in a membrane by doubling the frequency of operation however
in a bending mode radiator one would have to quadruple the frequency for the
same narrowing effect due to the square root increase in plate phase velocity.
However in this case the angle of the beam direction would move towards the
normal as the frequency increased. The overall radiation pattern is the
combination of the radiation pattern of the two waves and so may look like a
"lumpy" omnidirectional response. For a plate in which the wave velocity is
supersonic in both directions there are four travelling waves per mode and so
one would get a "bunch of bananas" type of polar pattern. As in the subsonic
case its shape and direction would be frequency dependent. For edge modes
which are a mixture of supersonic and subsonic propagation the resulting polar
pattern is a combination of the subsonic and the supersonic in the appropriate
axis.
In a practical radiator more than one mode would be excited at a given and so
the overall patter would be a superposition of the individual patterns and would
again look superficially omnidirectional. However there would still be rapid
changes in the detail of the polar pattern over small frequency changes, which
could compromise imaging. Also in general the loudspeaker would tend to get
more directional as the frequency increases because of the movement of the
main lobes to the normal in conjunction with it getting narrower.

Figure 19 The polar pattern of a supersonic mode in a strip radiator

16
 J

/ /
/ ·J / l

/ /

i _'J ·

X .... /

Figure 20 The polar pattern of a supersonic mode in a strip radiator of twice the
width

3 RESULTS
As seen earlier a panel radiator radiates either a comb or a "bunch of
bananas" radiation pattern depending on whether the panel is operating above or
below coincidence. Furthermore the angle of the main lobes depends on how
far greater than supersonic the plate velocity is. It is parallel to the plate if the
plate wave speed is equal to the air speed and approaches the normal direction
as the plate wave speed approaches infinity. This section looks at some
published results from an NXT panel [7]. As the NXT panel operates for most of
its ran below coincidence on would expect to see a figure of eight pattern due
to the comb radiation pattern. As the frequency approaches that of the
coincidence frequency on would expect to see lobes appearing at the sides of the
loudspeaker. This would give the impression of increasing dispersion at higher
frequencies. Figures 21 to 23 show a variety of polar plots of NXT panels.
They all show evidence of the edge beaming effect predicted, at a frequency of
about 4kHz. Although these polar patterns can be superficially interpreted as
diffuse radiation, in fact they are not. As described earlier the polar pattern
from such speakers is phase coherent and highly frequency dependent.
However, due to the rapid variation of the polar pattern with angle and its
sensitivity to dimension and frequency, some of the benefits of diffuse radiation
are obtained.

4 CONCLUSION
The Distributed Mode Loudspeaker's polar response has been analysed with
reference to it's radiation mechanism. In particular the polar response below and
above the supersonic plate propagation boundary has been examined. Below
the supersonic boundary a quasi omnidirectional response is possible. But all
response consist of the sum of multilobed patterns and hence do not provide
coherent radiation like a conventional radiator.

17
 .- 90

J 2()

'1

I_;(J ]0

//, {

t_o -' ' (;

;10

;00

27_
-- 250 tt_
' 1'-- 1000 Ih
"'--'"' 4000 Hz
_-_ 16000 IIz

Figure 21 A high resolution polar plot of an NXT panel. (from [7])

18
 170

120 60

150 3O

3OO

2/O
250Hz
--' 500Hz
-- lkHz
-- 2kHz
--- 4kHz
'-' 8kHz

Figure 22 A normalised polar plot of an NXTpanel. (from [7])

19
 B0 _ 80
.<-::"::i"-'-.-22..-- ':' .--'-':"-
.--' -_.-'" I "._ '.,.
...__:_/ .... :,., r : '- >< i ..../:"
_.:...f..
-_---_---;?,'_%-'_.--!-------!to
---_"-'"'!'"-_':_i'::
! '"'.'':"---o !._
,-'--_--._----_t':;----'---+--.----'t
'" -'"' '"_ >',- ."." t'." ',--%.'
_%'"--,.':
.-':-"1
__:,..-_'.:
:,.i,,.:..:,.._:_:._:...,:
"../" _'-,,._"';.,,.
..-:'/
:7 ..../:.k:_..._.',.f..
_/-y _.:-,-.;
'-.._
"._
':,_,..-..
"....
"-.,,."
...._" '-...'"
....' '_...'-¥'
....._" ',,..-'
:7 ,q:...
'?.."-_'...':..:.'.....'7
".......
:?':'
90 BO BO

.-'¥ ,...... Y'_ .K."_ , .i, '>-

-::.
.:F.'.
.."_, ,_,.
:._
T.-.'_- _:'"_'_'.,'Tt,,.

_..>
,..'.....:: :_::..,,_=,.: _,-?,.-i'-:_,:,_,"
'.,..:,_:_:,.-.,.:,.:
'_......¥ ..__ ._,
..' ..../
:"
.::'"-'""..... ,-: ... [(
..:,...,.,.;:,.'_.!::
....,...., ..._2..,....-...,
'-i..'
::.,;-
::.:.fi:
,....-:
,,--_,,,'::)
°
'.:..
_ _---.a.?_.
"..:.".._:. :.,.>,
a>>-'
'-¼:_:
__q .......
._,
.......
_o_:._>
:/
....
> _:.._:'
::::_::
_'-_';.
Figure 23 Measured polar plots of an NXT panel mounted in a ceiling.
(from [7])

8 REFERENCES
[l] Ver, I. L., and Holmer, C. I. In "Noise and Vibration Control", ed. L. L.
Beranek, pub McGraw-Hill, 1971, New York, pp 287-296.
[2] Pierri, R. A. "Study of a Dynamic Absorber for Reducing the Vibration
and Noise Radiation of Plate-Like Structures", M.Sc. dissertation, 1977,
University of Southampton, England.
[3] Fahy, F., "Sound and Structural Vibration: Radiation, Transmission and
Response", pub Academic Press, 1985, London.
[4] Waterhouse, R V, "Output of a Sound Source in a Reverberation
Chamber and Other Reflecting Environments" Journal of the Acoustical
Society of America, 27, no. 2, March 1955.
[5] Lord Rayleigh, "Work Done by Detached Sources", in "Scientific
Papers", pub Dover, New York, 1964, vols 5-6, pp135-141.
[6] "A Technical White Paper" pub New Transducers Ltd, 1996
[7] Peter Mapp and Martin Colloms, "Improvements in Intelligibility
through the Use of Diffuse Acoustic Radiators in Sound Distribution", Audio
Engineering Society 103rd Convention, 26-29 September 1997, New York,
USA, preprint #4634.

20