Engineering art: the science of concert
hall acoustics
TREVOR J. COX
Acoustics Research Centre, University of Salford, UK
PETER D’ANTONIO
RPG DiVusor Systems, Upper Marlboro, MD, USA
Modern concert hall design uses science and engineering to make an acoustic which embellishes and enhances the
artistry of the musicians. The modern discipline of concert hall acoustics is a little over a hundred years old, and
over the last century much has been learnt about how to ensure the audience receives high quality sound. During
this period, knowledge from a large number of disciplines has been exploited. It is the intention of this paper to
illustrate how disciplines as diverse as X-ray crystallography, psychology, and mobile telephony have in uenced
acoustic design. The paper will concentrate on the design of acoustic diVusers for concert halls, as this is a topic
currently attracting considerable interest within the acoustics industry and academia.
The acoustic of a concert hall contributes an important
speech sound intelligible, or to make music sound
part of the sound heard in a classical music perform-
beautiful. Modern acoustic science cannot guarantee
ance; the concert hall embellishes the sound. Music
a great acoustic every time, but if advice is followed,
outdoors may be popular when accompanied by re-
technical knowledge should ensure that bad halls are
works, but the quality of the sound is usually poor.
not built, while signi cantly increasing the chance of
Move indoors and the sound comes alive, enveloping
greatness being achieved. What makes a good concert
and involving the listener in the musicmaking process.
hall is a combination of many acoustic and non-
Outdoors, listeners receive sound straight from the
acoustic attributes perceived by an audience member
orchestra, there are no re ections from walls, and the
in a complex manner. The acoustic of a concert hall
sound appears distant. When music is played in a
might be perfect, but if the audience get soaked in
room, re ections from the walls, ceiling, and oor
rain walking from the car park, they are unlikely to
add reverberation and other characteristics to the
rate the experience highly. So a great design is about
sound. As the conductor Sir Adrian Boult said:1 ‘The
accommodating a wide range of requirements, and not
ideal concert hall is obviously that into which you
just acoustics. This article, however, will concentrate
make a not very pleasant sound and the audience
on the acoustics. When designing a hall, the acoustic
receives something that is quite beautiful. I maintain
engineer will look at many acoustic factors: the
that this really can happen in Boston Symphony Hall;
background noise level, the amount of reverberation
it is our ideal.’ This quote is of great historical
(the decay of sound after a note has stopped being
interest, because Boston Symphony Hall was the rst
played ), the amount of sound arriving from the side,
concert hall where the principles of reverberation
the re ections musicians receive that are necessary
were applied. These principles were developed about
for them to play in time and form a good tone, and
a hundred years ago by Wallace Sabine, who was
so on. The overall focus of this paper will be on the
the rst person to apply modern scienti c principles
role of surface diVusers.
to room design and pioneered modern concert hall
Currently there is much debate about the role
acoustics. Boston Symphony Hall is still seen as one
of surface diVusers in concert halls. To take two
of the greatest auditoria in the world. Conversely, a
examples, one eminent concert hall designer regularly
poor hall can have a detrimental eVect on the enjoy-
claims that too much diVusion is detrimental to the
ment of a performance, something that can aVect the
sound quality of the upper strings, while in contrast
musicians as well. As the conductor Sir Simon R attle
other acoustic engineers have blamed the disappoint-
said of one hall (which will remain nameless): ‘The
ing acoustics of certain major halls on a lack of
*** hall is the worst major concert arena in Europe.
surface diVusion. We will return to these contradictory
The will to live slips away in the rst half hour of
opinions later, and will try to shed some light on why
rehearsal.’2
they arise, but rst it is necessary to describe what a
Since Wallace Sabine’s work on room acoustics,
diVuser is and to outline the current state of the art
much has been learnt about what is important to get
a good acoustic, whether the requirement is to make
of acoustics.
INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2 119
© 2003 IoM Communications Ltd D OI 10.1179/030801803225010412
Published by Maney for the Institute of Materials, Minerals and Mining
1 Spatial and temporal responses of sound reflected
from a plane flat surface (above) and a diffuser
(below)
Treatments
To alter the acoustics of an existing room, treatment
2 Temporal and frequency responses from flat
is usually placed on the boundaries, for example if
(above) and diffusing (below) surfaces
an oYce is too reverberant or lively, absorbent ceiling
tiles or carpet might be used to absorb and so remove
some of the acoustic energy. In concert halls, the
faithful rendition of the original sound direct from
the instrument, and less coloration will be heard.
sound can be altered by placing treatment on the
walls and ceiling (the oor already has the audience
Figure 1 also shows a cross-section through a diVuser,
in this case a re ection phase grating consisting of a
and seating on it, and so is diYcult to alter). There
are three basic forms of treatment, large at surfaces,
series of wells of diVerent depths but the same width.
There are many types of diVuser, as explained below,
absorbers, and diVusers. Absorbers, such as the ceil-
ing tiles and carpet mentioned already, are not often
but in principle any non- at corrugated surface will
have some kind of diVusing ability. DiVusers can also
used in large concert halls, because they remove sound
energy from the space. In a hall every bit of energy
be constructed from a wide range of materials, such
as wood, gypsum, concrete, metal, and glass, the key
must be conserved because there is a limit to how
much energy an orchestra can produce. Consequently,
feature being that the material should be hard and
acoustically non-absorbent.
the designer must choose between at surfaces or
diVusers.
DiVusers are used in a variety of ways. For instance,
they can be used to reduce echoes arising from the
Figure 1 contrasts the spatial and temporal responses
of at and diVusing surfaces. These describe what the
rear walls of auditoria. Sound takes a long time to
travel from the stage to the rear wall of a concert
listener would receive if they were close to one of
the surfaces, and if no other surfaces were present.
hall, and if a strong re ection comes back from the
rear wall to the front of the hall, this may be heard
A at surface behaves like a mirror re ecting light,
the sound energy being preserved and concentrated
as an echo. In older halls the echo problem would
have been dealt with by placing absorbent on the rear
in the specular re ection direction, and with equal
angles of incidence and re ection. The time response
wall to absorb the acoustic energy, preventing the
re ection occurring and so curing the echo problem
shows the similarity between the direct sound and
the re ection: the at surface does little to the sound
(such a solution was adopted in London’s R oyal
Festival Hall ). However absorption removes acoustic
except change the direction in which it propagates.
F igure 2 shows the resulting frequency response. This
energy and so reduces the loudness of the orchestra.
A modern solution is therefore to use diVusers to
shows how the level (or volume) of the sound will
vary as the pitch (or frequency) of a note changes.
break up and disperse the troublesome re ections,
which can be done without loss of acoustic energy.
The frequency response of the at surface is uneven,
with a regularly spaced set of peaks and troughs, and
An example of the use of re ection phase grating
diVusers, on the rear wall of Carnegie Hall in New
is known as a comb lter response. This unevenness
means that some frequencies will be emphasised, and
York, can be seen in F ig. 3. In the UK, Glyndebourne
Opera House uses convex curved surfaces on the rear
some deemphasised. This leads in turn to coloration
of the sound, where the timbre of notes is altered.
wall to disperse sound.
A diVuser, on the other hand, disperses the re ection
both temporally and spatially. The time response
Architectural trends
is greatly altered, with re ections arriving over a
longer period, and the frequency response shows less
In older, pre-twentieth century halls, such as the
Grosser Musikvereinssaal in Vienna, ornamentation
evidence of comb ltering than the plane surface,
the peaks and troughs being uneven and randomly
appeared in a hall because it was the architectural
style of the day. Such walls were therefore naturally
spaced. This means that the re ected sound is a more
120 INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2
3 Schroeder diffusers (QRDs) applied to the rear
wall of Carnegie Hall to prevent echoes
diVusing; large at surfaces were very rare. The
Grosser Musikvereinssaal is an interesting example
to acousticians, because it is often cited as one of the
best halls in the world. The hall sound is thought to
have in uenced composers including Brahms, Bruckner,
and Mahler. In the Grosser Musikvereinssaal the
in uence of the surface diVusion on the sound is very
obvious, with a diVuse sound resulting.
In the twentieth century, however, architectural trends
changed and large expanses of at areas appeared
in many concert halls. The U K has many post-war
concert halls which have very little ornamentation,
such as the Colston Hall in Bristol. The style of the
day was to produce clean lines following a modernist
style, but these surfaces then had little or no diVusing
capability. The expanse of at surfaces can lead to
distortion in the sound heard as a result of comb
ltering, echoes, and other mechanisms. It is worth
noting, however, that it is also possible to design
very successful halls with at surfaces, a good UK
example being Symphony Hall in Birmingham, which
has relatively little surface diVusion.
The key to good diVuser design is to nd forms
that complement the architectural trends of the
4 Three different Schroeder diffusers: the original
day. The diVuser must not only meet the acoustic
design (top), a fractal design (middle), and an
speci cation, it must t in with the visual scheme
active diffuser (bottom); the diffusers are 0·6 m
wide, 0·6 m high and about 0·2 m deep
required by the architect. As discussed below, modern
diVuser designs have successfully been developed to
complement modern architectural forms.
Fowler Centre, New Zealand.4,5 Figure 5 illustrates this
application. Marshall and Hyde used large overhead
Schroeder diffusers
re ectors to provide early re ections to the audience
in the balconies in a revolutionary design. This was
The development of the modern diVuser began with
pioneering work by Manfred Schroeder, one of the
a layout whereby a hall could have good clarity, and
yet maintain a large volume for reverberation. The
twentieth century’s greatest acoustic engineers. In
the 1970s, Schroeder developed the phase grating
large volume partly comes from the space behind
the diVusers. Not many years before the design of
diVuser,3 also known as the Schroeder diVuser. An
example of the original design can be seen in F ig. 4
the hall, it had been established that lateral re ections
were important in concert halls as they promote a
(top). These diVusers oVered just what acoustic
designers were looking for, de ned acoustic perform-
sense of envelopment or spatial impression in rooms.6
The evidence for the bene cial eVects of lateral
ance based on very simple design equations; for while
it is know that ornamentation produces diVusion, it
re ections came from laboratory measurements on
human perception, which followed techniques pioneered
does this in an ill de ned and haphazard fashion.
One of the pioneering applications of Schroeder
in experimental psychology. These measurements
showed that lateral re ections are important to get a
diVusers was by Marshall and Hyde in the Michael
INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2 121
5 Schroeder diffusers in the Michael Fowler
Centre, New Zealand (photo courtesy Dr Harold
Marshall of Marshall Day Acoustics)
6 Scattered levels from a Schroeder diffuser
(left) and a plane surface (right) of the same
sense of involvement with the music. The need for
dimensions
lateral re ections in uenced Marshall and Hyde to
apply diVusers to the large overhead surfaces rather
than using at re ectors.
moves around the surface on a semicircle. A series of
lobes are seen, eleven in this case, which are grating
Another reason for this new diVuser technology
entering wide use was its commercialisation by an
lobes generated by the periodicity of the surface
structure. Imagine viewing this polar response end
American company, R PG DiVusor Systems Inc.,
whose interest lay in studio design. Around the time
on so that a set of eleven bright spots are seen; it is
this type of image that X-ray crystallographers use
that Schroeder was developing the new diVuser, a
new design regime for listening and monitoring rooms
to determine crystal structures. The problem posed in
the acoustic case is however somewhat diVerent from
was invented. This was the Live End Dead End
(LEDE ) layout,7 which was later re ned into the
the crystallographic challenge: in crystallography, the
diVraction patterns of the X-rays are used to deter-
R e ection Free Zone (R FZ ) design. DiVusers are
used in small spaces to disperse re ections that would
mine an unknown structure, whereas in the acoustic
case, the problem to be solved is how to determine
otherwise arrive early and at a high level and so cause
coloration of the timbre of the sound. This is some-
the correct surface structure to achieve a desired
polar response (or diVraction pattern). But before
times referred to as acoustic glare, and is again caused
by comb ltering. Just when studio designers were
explaining how Schroeder solved this problem, it is
necessary to explain how diVusers scatter sound.
looking for diVusers to achieve this design, by happy
coincidence Schroeder diVusers became available.
At that time, one of the founders of R PG, and
Huygens
also one of the authors of this paper, Peter D’Antonio,
was a diVraction physicist at the Laboratory for the
The Huygens construction used in optics is one way
of explaining how diVusing surfaces scatter, though it
Structure of Matter at the Naval Research Laboratory
in Washington, DC. Knowing of his interest in music,
is only approximate in many acoustic cases. Consider
a planar surface, the situation illustrated in the upper
a colleague handed him the latest issue of Physics
Today with a cover photo of Manfred Schroeder
half of F ig. 7. When illuminated by a sound source,
a set of secondary sources is generated on the surface,
seated in an anechoic chamber. The associated article
suggested using Schroeder’s number theoretic diVusers
and these are shown as stars in F ig. 7. Each of these
secondary sources then radiates semicircular waves.
in concert halls. It became apparent that the re ection
phase gratings suggested by Schroeder were in eVect
By connecting points on these waves which are in
phase with each other, it is possible to visualise the
two-dimensional sonic crystals, which scatter sound
in the same way that three-dimensional crystal lattices
waves that are re ected from the surface. (These are
rather like ripples on the surface of water created
scatter electromagnetic waves. Since the diVraction
theory employed in X-ray crystallographic studies
when a stone is thrown into the water.) In this
situation, a simple plane wave at right angles to the
was also applicable to re ection phase gratings, it was
straightforward to model and design the re ection
surface is generated. The planar surface is acting like
an acoustic mirror, and the wave is unaltered on
phase gratings using techniques rst developed in
crystallography.
re ection (except in its direction). The lower part of
Fig. 7 shows the case for a semicircular surface. In
F igure 6 ( left) shows the scattering from a
Schroeder diVuser in a polar response. A source
this case, the re ected waves are now semicircular in
shape. The wave has been altered by the surface, being
illuminates the surface, normal to the surface and
from the right. The polar response shows the energy
dispersed so that the sound re ects in all directions,
a characteristic desirable in a diVuser.
scattered from the surface (in decibels) as a receiver
122 INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2
method, Schroeder turned to his favourite subject of
number theory, at rst glance a rather obscure form
of abstract mathematics which studies the properties
of natural numbers, but which in practice has proved
to be very useful to scientists and engineers.
In the late eighteenth century, Carl Friedrich
Gauss developed the law of quadratic reciprocity,
well known to mathematicians working in number
theory. Dedekind was a doctoral student of Gauss
and wrote a ne description of his supervisor:8
… usually he sat in a comfortable attitude, looking down,
slightly stooped, with hands folded above his lap. He
spoke quite freely, very clearly, simply and plainly: but
when he wanted to emphasise a new viewpoint … then
he lifted his head, turned to one of those sitting next to
him, and gazed at him with his beautiful, penetrating
blue eyes during the emphatic speech. … If he proceeded
7 Huygens constructions for a plane wave reflected
from an explanation of principles to the development of
from a flat surface (above) and a curved surface
mathematical formulae, then he got up, and in a stately
(below): normal incidence source with incident
wavefronts excluded for clarity and secondary
very upright posture he wrote on a blackboard beside
sources shown as stars on the surface
him in his peculiarly beautiful handwriting: he always
succeeded through economy and deliberate arrangement
in making do with a rather small space. For numerical
Figure 8 shows the case for a simpli ed re ection
examples, on whose careful completion he placed special
phase grating. In this situation, the re ected waves
value, he brought along the requisite data on little slips
are delayed because the waves must travel down
of paper.
each well and back up again before re ection. The
Although best known to modern physicists for
secondary sources have diVerent delays (phases)
G auss’s Law, which explains the properties of the
because of the diVerent well depths, and this alters
electric eld, it is Gauss’s number theory work
the re ected wave. This again generates dispersion.
which is of most interest here, because it leads to
the quadratic residue sequence used in the design of
Sequences
the quadratic residue diVuser (QR D), an example
of which is shown in Fig. 4 (top). The formulation of a
In many ways, a re ection phase grating is acting
quadratic residue sequence is based on a prime number.
like an optical diVraction grating. In the acoustic
For the diVuser in F ig. 4 (top), the prime number is
case, the designer has control over the phases of the
7. The depth of the nth well is then proportional to
sound waves. To design a re ection phase grating, a
n2 modulo 7, where modulo indicates the smallest
method is required to determine an appropriate well
non-negative remainder. So the third well has a depth
depth sequence, which then generates a phase distri-
proportional to 32 modulo 7, in other words 2. The
bution on the surface of the diVuser to give the
sequence mapped out in this case is 0, 1, 4, 2, 2, 4, 1,
desired re ected wavefronts. In inventing such a
which can be seen in Fig. 4. (The quadratic residue
diVuser in F ig. 4 has zero depth wells on both ends,
but these are half the width of the others, a useful
sleight of hand to make manufacturing and tting
easier). If this quadratic residue sequence is used to
construct the diVuser, then the diVraction or grating
lobes generated all have the same energy, as shown
in Fig. 6 (left ).
There are many other sequences that can be used.
Another popular one is the primitive root sequence.
The depth of the nth well is then proportional to rn
modulo N , where r is a ‘primitive root’ of N and N
must be a prime. For r to be a primitive root, the
sequence generated must contain every integer from
1 to N 1 without repeat. Thus for N =7, r=3 is a
primitive root, which generates the sequence 1, 3, 2,
6, 4, 5. If this sequence is used to make a diVuser,
the central (specular) lobe will be suppressed, while
the others remain at the same level.
8 Huygens constructions for a plane wave reflected
This sequence generation technique is an advanced
from a simplified Schroeder diffuser: the upper
form of the number games sometimes played by
plot shows wavefronts from two wells only
for clarity
children: with some simple generation rules, fantastic
INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2 123
answers are produced. But these sequences are
seriously useful, and have been developed for diverse
applications in astronomy, in error checking systems
for computers, and in digital audio data and mobile
telephony. As Schroeder is fond of saying, number
theory is unreasonably useful, considering it was
developed as an abstract mathematical subject.
Improvements
While the basic Schroeder diVuser based on number
theory sequences is an ingenious invention, it has
several Achilles heels. Since the development of the
initial design, several revisions have been suggested
to improve performance. These are used to overcome
key weaknesses related to bandwidth, periodicity, and
appearance and will be discussed in detail below.
In concert hall acoustics, designs have to work
over a wide bandwidth. Human hearing extends over
between ten and eleven octaves (20 Hz to 20 kHz),
and diVuser designs are typically considered over
the seven most important (80 Hz to 5 kHz). This is
one of the key diVerences between much research in
acoustics and optics, as optics researchers are often
concerned with a single frequency or a narrow band-
width. In acoustics, the bandwidth is much wider. To
make a wide bandwidth diVuser, the wells need to be
narrow and deep, but this makes the device very
impractical: rst of all the structure becomes very
expensive to make, and second it becomes highly
absorbent (air is a viscous uid, and as with any such
9 The
Mandelbrot
set
at
two
different
uid it is diYcult to force into narrow wells, acoustic
magnifications
energy being lost in the process and converted to
heat). A solution to this problem has been developed,
inspired by chaos theory and fractals.
gratings require periodicity to work, many periods
of the diVuser are stacked next to each other. The
To handle many octaves, a diVuser needs to have
roughness on diVerent scales. The use of elements of
diVraction lobes are also a function of periodicity,
and to achieve even energy lobes (Schroeder’s
diVerent sizes is common in loudspeaker design. In
two way loudspeakers, for example, the large ‘woofer’
de nition of optimum diVusion) requires the structure
to be periodic. Yet these diVraction lobes represent
is used to radiate bass frequencies, and the smaller
‘tweeter’ generates the treble sound. For diVusers,
energy concentrated into particular directions, with
a lack of re ected energy between. A better diVuser
some roughness needs to have dimensions metres
in size, and some needs to have dimensions centi-
would be one that distributed the energy more evenly
in all directions without lobes. Consequently there is
metres in size. Fractals are objects which have
scaleable properties, and one of the most famous is
a contradiction: to use the original number theory
design, periodicity is required, yet this results in worse
the Mandelbrot set, shown in Fig. 9 at two diVerent
magni cations. When the set is greatly magni ed, by
performance. A solution to this paradox was devised
by Angus,10 who showed that techniques devised
around four thousand times in F ig. 9, a very similar
shape to the original is seen. The same eVect can be
in mobile telephony could be adopted for diVusers.
These techniques are also applied to the design of
achieved for diVusers, as shown in F ig. 4 (middle).
In the surface shown, smaller diVusers are mounted
loudspeaker and microphone arrays.
Code D ivision Multiple Access (CDMA ) systems
within larger ones, the smaller scattering the high
frequencies, and the larger the low frequencies. This
are used in mobile telephony to enable multiple users
to use the same transmission bandwidth. Of interest
type of diVuser is rather ttingly sold under the brand
name D iVractal.9 The example shown has diVusers
here are so called spread spectrum techniques. These
techniques take frequency (spectral ) components,
with two diVerent scales. Three diVerent sizes of
diVuser are needed to cover bass frequencies, the
and spread them over a frequency bandwidth. If the
lobes generated by the Schroeder diVuser are viewed
largest having a size about fty times greater than
the smallest.
as spatial frequency components, then when spread
spectrum techniques are applied, the lobes will be
The issue of periodicity is curious, as in many ways
it is the curse of the structure. Since re ection phase
spread spatially. This eVect is shown in Fig. 10, where
124 INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2
better to apply a sequence from number theory (yet
another case of number theory being unreasonably
useful ). This is particularly true if only a few periods
of the diVuser are being considered, as mathematicians
have produced methods for generating the best
possible sequences with the least amount of repetition.
Alternatively it is possible to task a computer to
laboriously search for the best sequences, but this is
rather slow. The rst sequence used for modulating
diVusers was the maximum length sequence, also known
as a Galois eld because of its basis in mathematics
developed by the nineteenth century mathematician
Evariste Galois. Galois unfortunately met an untimely
death in a duel at the age of twenty-one, but not
before he had sketched out some very important
mathematical concepts. As his director of studies
wrote:11 ‘It is the passion for mathematics which
dominates him, I think it would be best for him if
his parents would allow him to study nothing but
this, he is wasting his time here and does nothing
but torment his teachers and overwhelm himself with
punishments.’ Maximum length sequences are used
widely in digital systems; in acoustics they are best
known for producing eYcient measuring systems,
such as listening rooms, lters, and loudspeakers.
The appearance of Schroeder diVusers is an
important impediment to their use, especially given
current taste in architecture and interior design, which
tends to favour curves and more organic shapes.
With Schroeder diVusers the acoustic treatment
imposes a distinctive visual aesthetic, and while there
are architects who favour the argument that form
should follow function, most prefer to determine
their own aesthetic. If an architect thinks a diVuser
10 Scattered polar distribution from a periodic
looks ugly, it will not be used, however important
arrangement (light line) and a modulated
the treatment is to the acoustic design. Consequently,
arrangement (heavy line) of a quadratic residue
diffuser
there is a need for designs that complement modern
architectural trends. Figure 12 shows a modern diVuser
design on the ceiling of a cinema in Seattle. This is a
the spread spectrum process has enabled the scattered
curved diVuser designed to oVer a visual complement
energy to be redistributed from the three lobes in all
to the curtaining on the stage, while providing the
directions (all spatial frequencies).
required acoustic performance. The diVuser disperses
This idea can be applied to diVusers as follows.
re ections from the ceiling which would otherwise
Two base Schroeder diVusers are used. The rst is
colour the sound. To design this sort of diVuser requires
a standard Schroeder diVuser, the second a diVuser
a new methodology, and for this it is possible to use
which produces the same polar response only with
numerical optimisation. This is a method commonly
opposite phase. This is easy to form by changing the
used in engineering, for example to design parts of
well depth sequence; in fact the second diVuser is the
the space shuttle. Numerical optimisation does not
reverse of the rst. Figure 11 shows two base shapes
have the eYciency and elegance of number theory
arranged in a modulated array, in other words in a
design, but it is extremely eVective and robust.
random arrangement. The modulation sequence does
Numerical optimisation tasks a computer to search
not repeat, and so the diVusers are no longer periodic
for an optimum solution to a problem, for instance
and the periodicity lobes disappear. This produces a
it could look to optimise a car engine component
much more even polar response.
to minimise weight while ensuring suYcient strength
Whilst in principle it would be possible to ip a
coin to determine the modulation sequence, it is far
and durability. In the case of diVusers, the computer
11 Cross-section through a modulation scheme using an N=7 quadratic residue diffuser and its inverse
INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2 125
12 The Cinerama in Seattle, WA with a diffusing
ceiling (photo courtesy University of Salford
and Harris-Grant Associates)
looks for the surface shape which gives optimal scatter-
ing. The procedure works by iteration. The computer
starts by guessing some curved surface shape, and
the scattering from the surface is then predicted in
13 Optimised curved surface in the Edwina
terms of the polar response. The predicted polar
Palmer Hall, Hitchin, UK (photo courtesy Arup
response is rated for its quality in terms of a ‘ gure
Acoustics)
of merit’, which by a process of trial and error the
computer can try and minimise by changing the sur-
face shape. The process continues until an optimum
about on land ). As with biological populations, to
enable dramatic changes in the population of shapes,
design is found, which occurs when a minimum in
the gure of merit is determined. The search process
mutation is required. This is a random procedure
whereby a small probability is de ned of any gene in
is not completely random because this would be too
slow – fortunately mathematical algorithms have been
the child sequence being randomly changed, rather
than coming direct from the parents.
developed to allow the search to be done eYciently.
Currently the most popular approach, using a so
Selecting shapes to die oV can be done randomly,
with the least t (the poorest diVusers) being most
called genetic algorithm, models the problem as
an evolutionary process, using survival of the ttest
likely to be selected. In biological evolution, the ttest
are most likely to breed and pass on their genes, and
principles to carry out an eYcient search.
A genetic algorithm mimics the process of evolution
the least t are the most likely to die; this is also true
with an arti cial genetic algorithm used to design
that occurs in biology. A population of individuals
is randomly formed. Each individual is determined
diVusers. Using these principles, the tness of successive
populations should improve, and the process is con-
by its ‘genes’, which in this case are simply a set of
numbers which describe the curved surface shape.
tinued until the population becomes suYciently t
for the shape produced to be classi ed as optimum.
Each individual (or shape) has a tness value ( gure
of merit) that indicates how good it is at diVusing
F igure 13 shows an example of another optimised
curved surface. In this case a concave wall in a music
sound. Over time new populations are produced by
combining (breeding) previous shapes, and the old
practice room would have caused problems of focus-
ing if untreated. Concave walls focus sound to a
population dies oV. OVspring are produced by pairs
of parents breeding, and the oVspring have genes that
point in the same way that concave mirrors focus
light. In acoustics, focusing eVects are obvious in
are a composite of the parents’. The oVspring shape
will then have features drawn from the parent shapes,
whispering galleries, such as the dome in St Paul’s
Cathedral in London. F igure 14 shows the polar
in the same way that facial features of a child can
often be recognised in the parents. A common method
response for a concave wall, revealing that the
scattered energy level is much greater for the receiver
of mixing genes is called multipoint crossover. For
each gene, there is a fty per cent chance of the
at the focal point. In treating this focusing problem,
it would have been possible to add absorption on the
child’s gene coming from parent A, and a fty per
cent chance of it being from parent B.
wall to remove the re ections from the curved surface.
But this would have removed energy from one side of
If all that happened was combination of the
parents’ genes, then the system would never look
the orchestra, and these re ections are needed so the
musicians can hear themselves and their colleagues.
outside the parent population for better solutions
(the ‘ sh’ diVuser would never get lungs and walk
R e ections are needed for the musicians to keep in
126 INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2
technique exists for folding a sequence into a two-
dimensional array, a technique commonly referred to
as the Chinese R emainder Theorem.
An example of a Chinese remainder problem
was posed by Sun Tsu Suan-Ching in the fourth
century ad:13 ‘There are certain things whose number
is unknown. … Divided by 3, the remainder is 2; by 5
the remainder is 3; and by 7 the remainder is 2. What
will be the number?’ From this rather unpromising
start, a method of sequence folding can be generated
which has been used in coding systems, cryptology,
14 Scattering from a concave arc (light line)
and X-ray astronomy. Incidentally, one answer to
compared with an optimised curved diffuser
the above problem is 23. The mask shown in F ig. 15
(heavy line)
is a length 1023 maximum length sequence which has
been folded into a 31 by 33 array using the Chinese
time, form a good tone, and blend the overall ensemble
R emainder Theorem.
sound. The solution is therefore to use diVusers, as
The problem with maximum length sequences is
these remove the focusing eVect from the sound while
that they are devised for systems that are bipolar,
preserving the acoustic energy. Figure 14 illustrates the
consisting of +1s and 1s. The hybrid surface on
eVectiveness of the diVuser in dispersing the focused
the other hand produces areas of no re ection (0s)
sound.
and re ection (1s), and so is inherently unipolar. This
can be a problem when designing these diVusers.
Most electronic systems have bipolar capabilities, and
Absorption for diffusion
can produce signals of the opposite sign, but this
In recent years, interest has been returning to number
is not true of bre optic systems, and hence optical
theory to generate a diVerent kind of diVuser, the
sequences have been developed. Optical sequences
hybrid surface. Construction of a hybrid surface is
were developed for use in bre optical CDMA pro-
shown schematically in F ig. 15. It consists of a piece
cesses. Fibre optic CDMA sequences, where the light
of acoustic absorbent covered with a perforated mask,
intensity is either on or oV, cannot have cancellation,
the mask then being hidden from view by a thin piece
and hence use unipolar sequences. These sequences
of acoustically transparent cloth. Where there are
can be used to design hybrid diVusers.
holes in the mask, absorption is generated; where the
F igure 16 compares the scattering from a hybrid
sound strikes a solid part of the mask, re ection occurs.
absorber–diVuser with that from a plane surface. The
This then forms a surface which partially absorbs,
hybrid surface provides greater dispersion. This dis-
and any re ected energy is diVused because of the
persion can be improved even further if the hybrid
random arrangement of holes. The key to good dis-
surface is bent and made corrugated, as this breaks
persion is the arrangement of the holes, which is best
up the specular re ection component further. This
done following a pseudo random binary sequence with
type of design is proving to be very popular in studio
optimal autocorrelation properties ( least repetition).
control and listening rooms.
When the sequence has a 1, a hole is drilled in the
mask, when it has a 0, no hole is drilled. Any repetition
Active technology
in the sequence will lead to lobes, so sequences are
needed which are dissimilar from shifted versions
The nal technology to be discussed here is active
of themselves. Again, number theory can provide a
control technology. Active noise control has caused
whole range of sequences.
much interest in the last twenty years or so, but the
Angus12 started by looking at maximum length
application of this technology is not widespread
sequences, the same sequences which were used origin-
because the practical implementation is costly and
ally to modulate Schroeder diVusers. The problem is
problematic. It has, however, been used successfully
that maximum length sequences are just strings of 1s
in controlling ventilation, car, and aircraft noise.
and 0s, and what is needed for hybrid surfaces is
Attention has now been turned to whether an active
a two-dimensional array of numbers. Fortunately, a
diVuser can be generated. In active noise control,
the noise source is cancelled by generating a signal
from a secondary source of the same magnitude
but opposite phase. The waves from the noise and
secondary source then cancel each other out. Active
diVusers do something slightly diVerent: they start by
cancelling out a re ection, but then the secondary
source adds in an arti cial re ection which mimics
the characteristics of a diVuser.
F igure 4 (bottom) shows an artist’s impression of
15 Construction of a hybrid surface: porous
absorber (left), mask (middle), and cloth (right)
an active diVuser. Loudspeakers are placed at the
INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2 127
most successful halls are usually those where there
is a good rapport between the acoustician and
the architect, where the engineering facilitates and
embellishes the art.
As concert hall designs have been improved over
the century, attention has been focused on diVerent
aspects of acoustic design. Currently, there is much
interest in understanding the role of surface diVusers.
As this paper has shown, much is now known about
the design of diVusers, but many questions remain
unanswered. The most important question to answer
is how much diVusion should be applied, and where
diVusers should be used. While acoustic designers
have produced many innovative new designs, the
understanding of how and why to apply diVusers lags
behind and is still largely based on precedence.
In 2001, Manfred Schroeder attended the Inter-
national Congress on Acoustics in R ome. H e com-
mented on some photos of optimised curved surfaces,
saying we had produced diVusers which were beautiful.
We were attered by such a tribute from the pioneer
of modern acoustic diVuser design. Using optimisation
it is now possible to design simultaneously to given
visual and acoustic requirements – perhaps a case of
‘engineering art’?
16 Hemispherical polar balloons showing scatter-
Acknowledgements
ing from surfaces: hybrid surface (above),
plane hard surface (below)
Figure 1 is taken from t. j. cox and p. d’antonio:
A coustic A bsorbers and DiVusers; 2003, London,
Spon; Figs. 2, 6–8, 10, 14, and 16 from p. d’antonio
bottom of some of the wells. By changing how these
and t. j. cox: ‘D iVusor application in rooms’,
controlled loudspeakers respond to incident sound, it
A pplied A coustics, 2000, 60, 113–142; and Fig. 3 from
is possible to change the characteristics of the wells.
p. d
’antonio and t. j. cox: ‘Two decades of sound
For example, it would be possible to make a well
diVusor design and development part 1: applications
appear longer than it really is. But why would one
and design’, Journal of the A udio Engineering Society,
go to such lengths, as the active system with its
1998, 46, 955–976.
electronics and control structure is diYcult to develop
and expensive to implement? Active systems are
being investigated to gain additional diVusion at
low frequency, in a range where it is very diYcult to
Notes and literature cited
gain diVusion by normal (passive) means. To obtain
1. l. l. beranek: Concert and Opera H alls: H ow they
diVusion at low frequency is very diYcult because
Sound, 69–74; 1996, Woodbury, N Y, Acoustical Society
the size of acoustic waves (wavelength) becomes many
of America.
metres in size, which means that a diVuser should be
2. s. moss: ‘Things can only get better’, Guardian, 1999,
extremely large as the diVuser’s dimensions need to
23 July.
be comparable with the wavelength. Active diVusion
3. m. r. schroeder: ‘Binaural dissimilarity and optimum
is a method whereby the acoustic eld can be dis-
ceilings for concert halls: more lateral sound diVusion’,
turbed at a low frequency using a shallower device.
Journal of the Acoustical S ociety of America, 1979,
65, 958–963.
These devices, however, are very much in their infancy
4. a. h. marshall and j. r. hyde: ‘Some practical con-
and it remains to be seen whether a practical device
siderations in the use of quadratic residue diVusing
can be constructed.
surfaces’, Proceedings of the 10th International
Congress on Acoustics, Sydney, 1980, paper E7.3.
Summary
5. a. h. marshall, j. r. hyde and m. f. e. barron: ‘The
acoustical design of Wellington Town H all: design
Much has been learnt about the design of concert
development, implementation and modelling results’,
halls over the last hundred years. A little over a
Proceedings of the Institute of Acoustics, Edinburgh,
century ago, the design of a hall was based on a com-
1982.
bination of precedence and luck. Nowadays, acoustic
6. m. barron: ‘The subjective eVects of rst re ections in
science means that concert hall design involves a
concert halls – the need for lateral re ections’, Journal
of S ound and V ibration, 1971, 15, 475–494.
combination of precedence, engineering, and art. The
128 INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2
7. d. david and c. davis: ‘The LED E concept for the
10. j. a. s. angus: ‘U sing grating modulation to achieve
wideband large area diVusers’, A pplied Acoustics, 2000,
control of acoustic and psychoacoustic parameters
in recording control rooms’, Journal of the Audio
60, 143–165.
11. See www-gap.dcs.st-and.ac.uk/~history/Mathematicians/
Engineering S ociety, 1980, 28, 585–595.
8. See www-gap.dcs.st-and.ac.uk/~history/Mathematicians/
G alois.html.
12. j. a. s. angus: ‘Sound diVusers using reactive absorption
Gauss.html.
9. p. d’antonio and j. konnert: ‘The QRD diVractal: a
grating’, Proceedings of the 98th Convention of the
Audio Engineering Society, 1995, preprint 3953.
new one- or two-dimensional fractal sound diVusor’,
Journal of the A udio Engineering S ociety, 1992, 40,
13. d. wells: The Penguin Book of Curious and Interesting
Puzzles; 1992, London, Penguin.
113–129.
Trevor Cox is Reader in Acoustics at Salford University, UK.
H e graduated with a degree in physics from Birmingham
U niversity, U K in 1988 and went on to study auditorium
acoustics at Salford U niversity, where he was awarded his
doctorate in 1992. Between 1993 and 1995, D r Cox worked
at South Bank University, London. A large proportion of
Dr Trevor J. Cox
his research concerns diVusing surfaces: his use of numerical
Acoustics R esearch Centre
optimisation techniques has led to the worldwide appli-
U niversity of Salford
cation of innovative diVusing surfaces. This paper is based
Salford M5 4WT
on his Isambard K ingdom Brunel Award lecture given to
U K
the British Association for the Advancement of Science’s
t.j.cox@salford.ac.uk
Festival of Science held in Leicester in 2002.
Peter D ’Antonio received his BS degree from St John’s
U niversity in 1963 and his PhD from the Polytechnic
Institute of Brooklyn in 1967. D r D ’Antonio has specialised
in a wide variety of scienti c disciplines including spectro-
scopy, X-ray and electron diVraction, electron microscopy,
software development, and architectural acoustics. H e
carried out research in diVraction physics at the N aval
Dr Peter D’Antonio
R esearch Laboratory, Washington, D C for thirty years.
R PG D iVusor Systems Inc.
He is now president of RPG D iVusor Systems Inc., which
651-C Commerce D rive
was founded in 1983 to carry out basic research in room
U pper Marlboro
acoustics and to develop designs and innovative number
MD 20774
theoretic architectural surfaces, to enhance the acoustics of
U SA
critical listening and performance environments.
INTERDISCIPLINARY SCIENCE REVIEWS, 2003, VOL. 28, NO. 2 129