You are on page 1of 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/233688510

A Tutorial on Scattering and Diffusion Coefficients for Room


Acoustic Surfaces

Article  in  Acta Acustica united with Acustica · January 2006

CITATIONS READS

91 11,336

7 authors, including:

Trevor John Cox Peter D'Antonio


University of Salford Chesapeake Acoustic Research Institute LLC
212 PUBLICATIONS   2,979 CITATIONS    158 PUBLICATIONS   2,155 CITATIONS   

SEE PROFILE SEE PROFILE

Jean Jacques Embrechts Jin Yong Jeon


University of Liège Hanyang University
83 PUBLICATIONS   789 CITATIONS    355 PUBLICATIONS   3,723 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

How head-related transfer functions and ear canal impedances grow | HRTFs and ear canal impedances of children View project

Volumetric diffusers: a new paradigm for acoustic treatment View project

All content following this page was uploaded by Trevor John Cox on 17 June 2015.

The user has requested enhancement of the downloaded file.


ACTA ACUSTICA UNITED WITH
Vol. 92 (2006) 1 – 15
ACUSTICA Review Article

A Tutorial on Scattering and Diffusion Coefficients


for Room Acoustic Surfaces

T. J. Cox1 , B.-I. L. Dalenback2 , P. D’Antonio3 , J. J. Embrechts4 , J. Y. Jeon5 , E. Mommertz6 ,


M. Vorländer7
1
: Acoustics Research Centre, University of Salford, Salford M5 4WT, UK
2
: CATT, Mariagatan 16A, SE-41471 Gothenburg, Sweden
3
: RPG Diffusor Systems, Upper Marlboro, MD 20774, USA
4
: Department of Electrical Engineering and Computer Science, University of Liege B28, B–4000 Liege, Belgium
5
: Hanyang University, School of Architectural Engineering, Seoul 133791, South Korea
6
: Mueller-BBM, 82152 Planegg, Germany
7
: Institute of Technical Acoustics, Technical University Aachen, 52056 Aachen, Germany

Summary
There is a need to be able to quantify how sound is reflected from a surface, in terms of how it is spatially
dispersed. In recent years, diffusion and scattering coefficients have been developed to enable room acoustic
prediction models to be more accurate, and designers to evaluate the worth of diffusing surfaces. Two coefficients
have been, or are about to be, enshrined in standards. With two coefficients being standardised, there is great
potential for confusion. This paper is intended to bring together authors from different areas to explain where the
coefficients should and should not be applied. In addition, this tutorial paper will revisit the rationale behind the
coefficients and contrast the coefficient definitions and performance.
PACS no. 43.55.-n, 43.55.Ka, 43.55.Mc, 43.20.Ei, 43.20.Fn

1. Introduction Diffuse field methods have the advantage of quickly ob-


taining a random incidence coefficient, but are difficult to
In recent years, coefficients have been developed to char- predict. A measurement method for obtaining a random in-
acterise the scattering or diffuse reflections caused by sur- cidence scattering coefficient has been standardised by the
faces in rooms. These coefficients have been developed to ISO [1] based on the Mommertz and Vorländer technique
meet the needs of geometric room acoustic modellers, dif- [2, 3]. Free field methods are often more laborious mea-
fuser manufacturers and room designers. Unfortunately, surements to carry out, but can be readily predicted and
there does not appear to be one ideal coefficient which are used to characterize the uniformity of scattering. A free
meets the needs of all interest groups, and there are no dif- field method for a diffusion coefficient, based on the Cox
fusion or scattering coefficients currently in the literature and D’Antonio technique, has recently been published in
that do not have flaws in application. While on first exam- an Audio Engineering Society standard information doc-
ination it appears possible to produce a watertight defini- ument AES-4id-2001[4]. This methodology is now being
tion of a coefficient, detailed analysis reveals problems. In- considered by an ISO working group.
deed, two different coefficient definitions are, or are about With two coefficients being standardised, there is great
to be, enshrined in international standards. While this may potential for confusion. The intention of this paper is to
appear unsatisfactory, it should be remembered that room bring together different interest groups to explain where
acoustics has used an absorption coefficient for a century the coefficients should and should not be applied. In ad-
which has well defined limitations in application. For ex- dition, this tutorial paper will recap the rationale behind
ample, there are two primary techniques for measuring ab- the coefficients and contrast the coefficient definitions and
sorption - the impedance tube and reverberation chamber performance using new and old measurements and predic-
methods - and each measurement method has advantages tions.
and disadvantages and is used for different reasons. In an
analogous manner, the methods for characterising diffuse The terms scattering and diffuse reflection are used in
reflections can be classified either as free or diffuse field. different ways and are interchanged in different subject
fields. (Often diffuse reflection is termed diffusion, which
can then be confused with volume diffusion.) Conse-
Received 12 July 2004, revised 21 june 2005, quently, to reduce misunderstanding, this paper will adopt
accepted 18 July 2005. the nomenclature that has been used in the standards. A


c S. Hirzel Verlag · EAA 1
ACTA ACUSTICA UNITED WITH ACUSTICA Cox et al.: Scattering and diffusion coefficients tutorial
Vol. 92 (2006)

scattering coefficient is a measure of the amount of sound law [11], where the diffused energy is proportional to the
scattered away from the specular reflection direction. A cosine of the angle of reflection. Lambert’s law is used be-
diffusion coefficient measures the quality (in terms of spa- cause it fits with the philosophy of the geometric models
tial uniformity) of reflections produced by a surface; in which are based on high frequency modelling techniques.
the case of the AES coefficient by measuring the similar- As shall be discussed below, diffusion coefficients used by
ity between the scattered polar distribution and a uniform diffuser designers are based on uniform energy distribu-
distribution. tion, where the reflected energy is uniform in all directions.
Uniform energy distribution is possible because the re-
2. The need for coefficients flected sound from surfaces display coherent interference
effects at the most important acoustic frequencies, indeed
2.1. Geometric room acoustic models this coherence is explicitly exploited in many diffuser de-
signs.
Originally, geometric room acoustic models did not in-
clude the effects of scattering generated by edges and sur- 2.2. Diffuser manufacturer and application
face roughness. In recent decades, however, considerable
evidence has been produced to show that incorporating The development of the modern diffuser can be traced
scattering into these models enhances prediction accuracy, back to the 1970s and Schroeder’s revolutionary designs
and in many cases is an essential ingredient in an accu- that try to achieve diffuse reflection [12]. Before Schroeder
rate model. For instance, it has been shown that without diffusers, there was very little or no information on how to
surface scattering, these room models tend to over predict create defined acoustic scattering. With the introduction
reverberation time [5, 6, 7, 8]. This is especially true in of surfaces like the Quadratic Residue Diffuser (QRD),
spaces where absorption is unevenly distributed, as hap- for the first time a measure of complete diffuse reflec-
pens in many concert halls, or where rooms are highly dis- tion was given. In this case, complete diffuse reflection
proportionate, as happens in many factories. Moreover, for is when the level of the grating lobes produced by peri-
acoustic parameters that are highly dependent on early re- odic phase gratings all have the same energy. Schroeder
flection prediction accuracy, such as early lateral energy diffusers have since been used in a wide variety of spaces
fraction and clarity, there can be great sensitivity to the [13, 14] and a few examples are given as follows. The in-
diffuse modelling technique and correct scattering coeffi- vention of Schroeder diffusers coincided with the develop-
cients [9]. In the first round robin study of room acoustic ment of small room design concepts, involving temporal
models [8], three models were found to perform signif- reflection free and diffuse zones [15, 16]. Temporal reflec-
icantly better than others. These three models produced tion free zones around the listening position are created by
results approximately within one subjective difference li- absorbing or diffusing the first order reflections. Diffusers
men, while the less successful models produced predic- are used on the rear wall to provide uniform, enveloping
tions inaccurate by many difference limen. What differen- diffuse reflections, creating passive surround sound. The
tiated the three best models from the others was the in- use of diffusers can reduce coloration effects due to strong
clusion of a method to model surface scattering. Including early reflections that naturally occur in small rooms. For
frequency dependent diffuse reflection is one of the neces- this reason, there is widespread use of Schroeder diffusers
sary conditions for good prediction accuracy in geometric in small critical listening environments. Other popular ap-
room models. plications for diffusers include the rear walls of large au-
There are many different methods for incorporating dif- ditoria to prevent echoes, and around stage areas to reduce
fuse reflections into geometric room models [10]. This coloration due to early arriving reflections and so aid en-
process is inevitably approximate, because the room mod- semble and the ability of musicians to produce a good
els can not explicitly represent the wave nature of sound. tone. However, some so-called diffusers do not actually
The room models use a scattering coefficient to determine diffuse in the way the designers intended, due to poor de-
the proportion of the reflected energy that is specularly re- sign [17]. The recent proliferation of diffusers appears to
flected, and the proportion that is scattered. (In the com- be a reaction to the dominant mid- late twentieth century
puter models these are often referred to as diffusion coef- architectural forms. Large flat surfaces became common
ficients, although some programs have already made the place and these generate strong specular reflections. Some-
change in nomenclature to scattering coefficient. The term times this can cause problems of coloration and echoes,
scattering coefficient is used here to be consistent with which can be corrected by absorbers or diffusers. There is
other sections of the paper.) Problems arise because un- a need for contemporary scattering surfaces that comple-
til recently there has not been a procedure for determining ment contemporary architecture in the way the statuary,
the values of the scattering coefficients for room models, coffered ceilings and relief ornamentation complemented
except by trial and error and through precedence. Conse- classic architecture.
quently, a key driving force behind the ISO process was to Since the 1970s, many other types of diffusers have been
standardise a method to enable random incidence scatter- produced, and to enable the merits of these designs to be
ing coefficients to be determined in a rigorous manner. evaluated, it is necessary to have a measure of the qual-
The scattered energy in a geometric model is usually ity of the diffuse reflections. The idea of measuring the
assumed to be distributed according to Lambert’s cosine evenness of grating lobes, which came from Schroeder’s

2
Cox et al.: Scattering and diffusion coefficients tutorial ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 92 (2006)

sured or predicted according to AES-4id-2001. Without


standardisation, the industry is vulnerable to published
performance data, which have no basis in fact and diffusers
incident (1-a)(1-s)
1 scattered that do not perform as intended. The evaluation criteria
(1-a)s developed do not just have to be applied to especially de-
specularly signed surfaces, they can also be used to monitor the dif-
reflected
energy fusion by “accidental” diffusers. It appears that diffusing
surfaces are often applied in a haphazard fashion because
there is an incomplete understanding of when and where to
apply diffusers. For instance, discussions with consultants
will produce examples where it is claimed that too much or
rough surface
too little surface diffusion resulted in acoustic aberrations.
A priory to developing a better understanding of where
Figure 1. Illustration of the separation of reflected energy into diffusers are needed, is an index to measure their quality.
scattered and specular components (After Vorländer and Mom-
mertz [3]).
3. Scattering coefficient

original work, is not a sustainable criterion because sur- 3.1. Intentions and applications of scattering coeffi-
faces do not have to be periodic and so did not neces- cients
sarily have grating lobes. A new paradigm based on uni- Some sound phenomena can be sufficiently described in
form scattering in all directions or in specific desired direc- terms of energy or particle radiation as well as a more com-
tions is needed. Consequently, new definitions to measure plex wave approach. This kind of dualism is not only use-
the diffuseness of reflections have been developed. This ful to the fields of light propagation or particle physics, but
has culminated in the publication of the Audio Engineer- to sound, too. Typically energy or particle models are used
ing Society standard information document AES-4id-2001 when broadband, short wavelength sound is considered.
[4], which gives measurement and calculation procedures For broadband sound problems and impulse responses, the
for a diffusion coefficient. For a diffusion coefficient to be coherence length of a pulse must be taken into account.
useful to designers, the primary characteristic of the coef- A reflection as a component of room impulse response is
ficient is that it must rank diffusers correctly according to rather short. In these cases, the scattered field cannot be
quality. This will not necessarily be achieved by the scat- easily calculated by coherent techniques like Fourier syn-
tering coefficient, as discussed below, and this is why a thesis or similar techniques. Accordingly interpretation of
separate coefficient has been developed for quality. the polar distribution of the scattered sound is not neces-
Current state of the art in diffuser design is numerical sarily useful, since the details of the distribution, the lobes
optimisation [18, 19]. This is an iterative design process: a and nulls, are spatially smeared with incoherent excita-
starting surface topography is generated, the scattered po- tion. Instead, an adequate quantity for describing broad-
lar response for the surface predicted, the quality of the band scattering on rough surfaces is a scattering coeffi-
polar response evaluated in a single figure of merit, then cient.
the surface topography is altered to improve the figure of The scattering coefficient is defined as the ratio of the
merit. This loop is repeated until a maximum in the figure non-specularly reflected sound energy to the totally re-
of merit is found. A key element of this design process is flected energy as shown in Figure 1. Although it is just
the figure of merit, and for this the diffusion coefficient is a rough single number, the coefficient is well adapted
used. Using diffusion coefficients in optimisation has en- for use in geometric room prediction methods involving
abled designs to move away from the rigid geometric con- high frequency modelling and scattered energy following
structs imposed by phase grating diffusers. This has en- some probability function. The coefficient is used to de-
abled designs where both acoustic and visual requirements cide whether a reflection is diffuse or specular in a room
can be considered and their conflicting requirements re- model. The diffuse reflection’s spatial distribution may
solved. It is now possible to make diffusers which blend in then be Lambert’s law, or any other distribution depend-
with architectural forms rather than appearing as add-ons; ing on the energy balance between specular and scattered
this is important for diffusers to be acceptable to archi- parts and on the angle of incidence [20]. In this sense,
tects. Optimized surface shapes can be fully reflective or the scattering coefficient has the same importance as the
partially absorbing as desired. absorption coefficient. Like the absorption coefficient, the
When a designer requires absorbing surfaces in a space, scattering coefficient varies with frequency and so is given
a performance specification in terms of the absorption co- in one-third octave or octave frequency bands.
efficient will be used to ensure quality and compliance
with design requirements. One of the aims of research into 3.2. Scope
diffusion coefficients was to facilitate the use of defined Although wavelengths are not physically consistent with
scattering ability in performance specifications; this can the incoherent energy approach, frequency dependent en-
now be done by specifying diffusions coefficients mea- ergy absorption and energy scattering models can predict

3
ACTA ACUSTICA UNITED WITH ACUSTICA Cox et al.: Scattering and diffusion coefficients tutorial
Vol. 92 (2006)

coefficient because the energy maybe scattered away from


the specular reflection direction, rather than just being con-
verted into non-acoustical energy. From these equations
the scattering coefficient s can be determined by:
h αspec − α
s= . (2)
a 1−α
The “specular absorption coefficient” directly corresponds
f << c/2a f » c/2a f >> c/2a
to the absorption coefficient if s = 0.
or and and From the general definition in equation (2), several
f << c/2h f » c/2h f >> c/2h
methods can be used to evaluate the scattering coeffi-
cient. All these methods differ in the way they separate
Figure 2. Frequency ranges for scattering from a periodic surface the “specular” and the “diffuse” energy. These include:
of repeat distance a, and roughness depth h. 1. The random incidence scattering coefficient measured
in the reverberation chamber [2, 3] by phase-locked av-
eraging of many room impulse responses as the surface
various phenomena of sound fields in rooms. Measure-
is moved by rotation. This is the “ISO” method [1] and
ments of incident and absorbed intensity with broadband
is given the symbol s.
signals can yield information about spatial and frequency
2. The free-field scattering coefficient obtained by phase-
averages. These averages are often sufficient in the pre-
locked averaging of several free-field responses as the
diction of noise levels, reverberation times or other single
surface is moved (e.g. rotated). Two approaches have
numbers related to octave or one-third octave bands. This
been suggested, one where the averaging is in the spec-
is, again, the typical large room assumption which is the
ular reflection direction only [2, 3], and another where
basis for classical room acoustics.
the averaging is done for receivers over a complete
The scattering coefficient generally depends on the fre-
hemisphere [22]. Both can produce a free-field scatter-
quency and angle of sound incidence. Similar to the ran-
ing coefficient at specific angles of incidence, which
dom incidence absorption coefficient obtained in rever-
can be integrated using Paris’ formula to estimate a
beration rooms, an angular average of the scattering co-
random incidence value.
efficient, i.e. the random incidence scattering coefficient,
3. The free-field correlation scattering coefficient ob-
can be defined as well. While in models of classical room
tained from the free-field polar distribution [23]. This
acoustics the room dimensions must be large compared
is obtained by comparing the polar response from the
with average wavelengths, the surface roughness must be
test surface with that for a reference flat surface. This
smaller than the average wavelengths for the scattering
can produce a free-field scattering coefficient at a spe-
models, see Figure 2. Both period width and roughness
cific angle of incidence, δc (θ), and this can be inte-
height must be considered.
grated using Paris’ formula to estimate a random inci-
In the light of the arguments above, broadband sound
dence value.
and incoherence, and the fact that room models wish to
It is also possible to take more pragmatic approaches to
separate out specular and diffuse components, the more
fit measured polar response data to the output of computer
appropriate quantity to be used in simulation software is
models of reflection [24, 25] and therefore bypass the need
the random incidence scattering coefficient rather than the
for separating out the specular and diffuse energy during
uniformity diffusion coefficient. Then it also follows that
measurement. But these methods (mostly) give values that
Lambert’s law, as statistical limit of many reflections, is a
depend on the room model chosen to process the measure-
reasonable approach [21], remembering that ray tracing is
ments, and so are not useful for standardization.
best with a large numbers of reflections and the late part The measurement of scattering coefficients using phase-
of impulse responses, rather than the early reflections with locked averaging is easiest to explain in the free-field. The
specific angular dependent behaviour. specular absorption coefficient is found by rotating the test
sample while phase-lock averaging the reflected pulses.
3.3. Rationale
Phase-locked averaging refers to the fact that the start of
The energies of reflections (normalised with respect to a the impulse responses must be correctly aligned to the
reflection from a rigid reference plane) in Figure 1 are ex- same point in time. Figure 3 shows three band pass fil-
pressed in terms of: tered pulses for different orientations of a corrugated sur-
face. The initial parts of the reflections are highly corre-
Espec = (1 − α)(1 − s) ≡ 1 − αspec , lated; this is the specular component of the reflection, and
Etotal = (1 − α), (1) remains as the sample is rotated. In contrast, the later parts
of the reflected pulses are not in phase and depend strongly
where s is the scattering coefficient, α is the absorption on the specific orientation; this is the diffuse component.
coefficient, Espec the energy specularly reflected and Etotal By averaging the reflected pulses while rotating the sam-
the total reflected energy. The coefficient αspec is the “spec- ple, the diffuse components are averaged to zero, and only
ular absorption coefficient”. This is an apparent absorption the specular energy remains.

4
Cox et al.: Scattering and diffusion coefficients tutorial ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 92 (2006)

Pressure

Time

Diffusion or scattering coefficient


Figure 3. Band limited reflected pulses for different sample ori-
entations (After Vorländer and Mommertz [3]).

Table I. Measurement conditions for the reverberation time mea-


surement.
Reverberation time Test sample Turntable

T1 not present not rotating


T2 present not rotating
Figure 4. Top: The polar response for a plane surface and the
T3 not present rotating
plane surface tilted. : Reference plane surface; ——: Plane
T4 present rotating
surface tilted by 10◦ . Bottom: The diffusion and correlation scat-
tering coefficient spectra for surfaces. : Diffusion coefficient,
plane surface – – –: Diffusion coefficient, tilted plane surface
Following procedures similar to those described in ISO ——: Correlation scattering coefficient, tilted plane surface (Af-
354 [26], the absorption coefficients α and αspec can be de- ter Cox and D’Antonio [14]).
termined by measuring four reverberation times as shown
in Table I. The necessary formulations are given in [1]. Fi- data for computer models in order to improve the accuracy
nally, the random-incidence scattering coefficient s is cal- of the results.
culated according to equation (2). The crucial difference
between the standard test method of absorption and the 3.5. Difficulties and discussions
test method for random incidence scattering is that addi-
tional reverberation times are to measured, one while the According to its definition, the scattering coefficient con-
empty turntable is rotating (T3 ), and one while the sample tains only limited information about the spatial distribu-
is turned (T4 ). Therefore a circular sample shape is the best tion of the reflected sound. Hence, it does not necessarily
choice due to the invariance of the edge diffraction against differentiate between redirection and dispersion. For in-
turning. Provided the surface is not so deep so that edge stance, for a slightly tilted flat surface (see Figure 4) or
diffraction is insignificant, only the contribution from the a concave focussing surface (see Figure 5), high values
surface corrugation causes decorrelation during rotation; for the correlation scattering coefficients can be obtained.
this is a measure of scattering due to roughness, not due to The correlation scattering coefficient interprets the redi-
edge diffraction. rection or focussing of the sound as being scattering and
therefore gives a high coefficient value at high frequencies.
The diffusion coefficient gives similar values whatever the
3.4. Advantages
surface orientation for the tilted surface, and interprets the
A main advantage of the method is that the scattering co- focussing surface as being worse than the plane surface.
efficient has a clear physical meaning. Its definition and This is an illustration of the diffusion coefficient being a
measurement using the ISO random incidence method is more strict measure of quality. The scattering coefficient
closely related to the definition and measurement of the only depends on energy moved away from a particular
absorption coefficient according to ISO 354. At model direction, and not on how this scattered energy is redis-
scale, even the time for the measurement itself is com- tributed. Although this may at first appear strange, it is
parable to the time required for the measurement of ab- consistent with the definition of the scattering coefficient
sorption coefficient. However, more recent studies [27, 28] as used in computer models. For example, if a saw-tooth
have shown that it is difficult to obtain fast measurements shaped surface as shown in Figure 2 is modelled in a room
in full-sizes reverberation chambers, although, the method model as a single plane surface, the use of a scattering co-
still gathers a random incidence value in a shorter time efficient considers that the sound at high and at mid fre-
than a free field method could achieve. As already ex- quencies are not specularly reflected with respect to the
plained, the scattering coefficient may be used as input modelled surface. If it is desirable for the geometry of the

5
ACTA ACUSTICA UNITED WITH ACUSTICA Cox et al.: Scattering and diffusion coefficients tutorial
Vol. 92 (2006)

Scattering / diffusion coeff.

Figure 6. Random incidence ISO scattering coefficient for an


empty turntable (diameter 600 mm, size of the reverberation
room 1 m3 ).

Figure 5. Top: Polar responses for a flat surface and a con-


cave sample. : Reference plane surface; ——: Concave sur-
face. Bottom: Diffusion and correlation scattering coefficients for
these samples. : Diffusion coefficient, plane surface, - - - - -:
Diffusion coefficient, concave surface; – – –: Correlation scatter-
ing coefficient, concave surface (After Cox and D’Antonio [14]).

Figure 7. The changes in the tails of two impulse responses due


surface be explicitly used to scatter sound energy, then it to time variance.
would be necessary to model the surface in more detail for
higher frequency bands. Therefore, the scattering coeffi-
The measurement principle of the ISO random in-
cients should not be used to interpret single surface items,
cidence method relies on the assumption that the de-
but should only be used for large surfaces with roughness.
correlation during the rotation of the sample is only due
A surface that does not diffuse uniformly but merely to the scattering properties of the surface. However, de-
redirects can be useful to prevent lock-in of sound between correlation also occurs due to small temperature devia-
walls etc., but they need to be used with great care to pre- tions during the measurement period or when the base
vent problems. For example, if used to reduce echo prob- plate of the turntable is not absolutely perpendicular or
lems, they may just move the echo from one place in a axisymmetrical with respect to the axis of rotation. This
room to another. is a particular logistical problem for full-scale measure-
The value obtained for the concave surface depends on ments [27, 28]. The error generally increases with increas-
the measurement process used to gain the scattering coef- ing frequency and can be easily checked by a measurement
ficient. Using the correlation scattering coefficient, which with an empty turntable, see Figure 6. According to the
derives the coefficient from a polar response, high scat- ISO standard [1] a measurement with an empty rotating
tering coefficient values are obtained at high frequency turntable shall be considered within the evaluation process
as shown in Figure 5. A concave surface centred on the to help compensate for this.
turntable using the ISO random incidence method will, Recent work on full scale measurements [27, 28, 29]
however, (by symmetry) lead to a scattering coefficient of have indicated that swept sine wave measurement may
zero. There are therefore different conclusions, depending work better at full scale than Maximum Length Sequence
on the method which is used to evaluate the scattering co- (MLS) techniques because swept sine waves deal better
efficient. Therefore, the methods for evaluating the scat- with time variance. Figure 7 shows parts of two impulse
tering coefficient should be regarded as different practical responses measured at the same position in the room, with
ways of gaining the coefficient, defined by a theoretical a fixed turntable and diffuser, after opening and closing
separation between specular and diffusely reflected energy. the entrance door, as a function of the number of samples
In the case of a surface which is randomly rough, it might (at 44.1 kHz) after the direct sound. These two impulse re-
be expected that the different measurement methods will sponse are phase-locked on the direct sound (not shown on
give similar results; for other surfaces, this will not neces- the figure). The delay between both impulse responses in-
sarily be true. creases with time (measured after the direct sound) and

6
Cox et al.: Scattering and diffusion coefficients tutorial ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 92 (2006)

this could be interpreted by a change in sound speed,


and therefore by temperature variation of about 0.01 ◦ C.
Small changes in temperature can induce significant ef-

Scattering coefficient
fects which are interpreted by the method as decorrela-
tion, and therefore scattering. These effects are not only
observed when opening and closing the door of the rever-
beration room, but also (but to a lesser extent) when the
room is kept closed, probably due to unavoidable insta-
bility of the propagation medium. As measurement time
grows (which is inevitable in real-size rooms), these time
invariances are more likely to influence the scattering co-
efficient; the effect is most notable at high frequencies
[30, 31]. Air movements created by the rotation of the
turntable does not seem to create short-time effects, i.e. Figure 8. Random incidence ISO scattering coefficients of hemi-
immediate variations in the impulse responses and rever- spheres on •: a circular area; 2: a square area; : and square plate
beration times. So, it seems that long-time instability of (1.6 cm high). (After Gomes et al. [34]).
the medium (possibly also influenced by air movements)
is the significant effect.
De Geetere and Vermeir [28] reported on the use of a square plate (1.6 cm high) placed beneath the hemi-
MLS and sine sweep signals and indicated that the best spheres, so that four edges were “inserted” in the mea-
signal to noise ratios were obtained using sine sweeps. An surement. The results are shown in Figure 8. There are
explanation has been proposed by Stan et al. [32]; the opti- practically no differences between the ISO scattering co-
mum output signal in the MLS technique is limited by the efficients measured from hemispheres spread over a circu-
need to reduce nonlinearities introduced by the measure- lar or over a square area. However, the additional edges,
ment system (mainly in the loudspeaker). By construction, which were later inserted in the measurement when a
the log-sine sweep method is less influenced by harmonic square plate was used, lead to excessively high values.
distortion, and therefore can obtain better signal to noise These results exemplify how inaccurate scattering coeffi-
ratios [33]. cients can be, if they are measured from square samples
Since the measurement method is intended to measure which contain edges.
surface roughness, the results are only reliable when the In order to minimise this kind of edge effects it is pos-
structural depth of the sample is small compared to the sible to hide the edges. In this measurement method, sam-
size of the specimen (approx. h ≤ d/16, where d is the ples which are symmetrical to the axis of rotation are ‘in-
diameter of the turntable and h the structural depth; these visible’. Similarly, circular edges which are symmetrical
limits are based on the measurement findings from a num- to the axis of rotation provide no or small influence over
ber of authors). If variations of the structural depth along the results. This can be approximated for square samples
the edges of the samples are too large, this may lead to by using a new base plate with a square recess within
scattering coefficients which are significantly higher than which the sample is placed [29].
those of an extended surface. This edge effect may occa-
sionally cause measured scattering coefficients larger than
1 and limits the measurement method to surfaces which
4. Diffusion coefficient
scatter at mid and high, rather than at low frequencies.
4.1. Rationale
Scattering coefficients greater than one are a problem in
room acoustics programs; in fact this is the same problem The diffusion coefficient is intended to be a measure of
generated by absorption coefficients greater than 1; these quality, as the primary goal in developing the coefficient
values must be modified by the user of the program in or- was to enable the worth of surfaces to be determined. The
der to be accepted as valuable inputs. Alternatively, differ- concept is to measure the uniformity of the sound scat-
ent sample scales have to be used [29]. So far, it is not clear tered from the test surface in an analogous way to how one
if a border around the sample may improve the measure- might test the uniformity of radiation from loudspeakers.
ment. While the ISO standard will contain the requirement To do this, the energy reflected from the surface is mea-
that a test specimen shall be circular to guarantee that edge sured or predicted as a free field polar response, examples
effects have no or only a small influence over the results, of which can be seen in Figure 9 [35]. The diffusion coeffi-
more recent work has shown that sometimes, it is possible cient is then a single number which gauges the uniformity
to use non-circular samples as mentioned below. of the polar response. If the same energy is scattered in
To quantify the influence of the edges, hemispheres all directions, then the diffusion coefficient is one; this is
were randomly placed on a circular and also on a square termed complete diffusion. If all the energy is scattered in
area (directly on the base plate) and the ISO random in- one direction, then the diffusion coefficient is zero. The
cidence scattering coefficients were measured [34]. One diffusion coefficient is evaluated in one-third octave bands
additional measurement was also performed, now with and is frequency dependent.

7
ACTA ACUSTICA UNITED WITH ACUSTICA Cox et al.: Scattering and diffusion coefficients tutorial
Vol. 92 (2006)

Consequently, there are two key stages to determining


the coefficient, the determination of the polar responses of
the reflected energy, and the data reduction to a single fig-
ure. The data reduction from a polar response to a single
figure of merit has attracted most attention in the literature.
There have been various statistical operations suggested:
standard deviation [18, 19, 36, 37], directivity [38], spec-
ular zone, spherical harmonics [39] and autocorrelation
[40]. Of these, AES-4id-2001 has favoured the autocor-
relation diffusion coefficient as being the most robust and
useful measure – see reference [40] for a detailed justifica-
tion. For a fixed source position, the diffusion coefficient,
dψ , can be calculated from:
n 2   2
10Li /10 − ni=1 10Li /10
i=1
dψ =   2
(n − 1) ni=1 10Li /10
n 
Li /10 2
i=1 10
≈   2 , (3)
n ni=1 10Li /10

where Li are a set of sound pressure levels in decibels, n


is the number of receivers and ψ is the angle of incidence.
This equation is only valid when each receiver position
samples the same solid reflection angle; AES-4id-2001 in-
cludes formulations for dealing with other cases.
The polar responses are obtained either by measure-
ment, say by MLS, or by prediction, say by Boundary El-
ement Methods (BEMs). AES-4id-2001 sets out the de-
tailed requirements such as the source and receiver posi-
tions and these will not be repeated here. Measurements
will usually carried out at a model scale, say 1:5, to enable
the source and receiver positions to be small enough to fit
in typical test rooms. The use of MLS enables the reflected
sound to be isolated from the incident sound by time gat-
ing [40, 41] and from a set of gated impulse responses the
scattered polar response is obtained. Figure 9. Top: A 3D polar response measured from a diffuser
The diffusion coefficient is intended to be evaluated ei- shown in the middle of the figure (After Cox and D’Antonio
ther in a single plane or over a hemisphere. Diffusers gen- [14]). Bottom: Polar responses for two single plane surfaces:
erally break down into two categories; (i) single plane ——: An optimized curved surface, and : a semi-cylinder.
diffusers, also known as 1D diffusers, where the surface (After Cox and D’Antonio [35]).
is extruded in one direction and the scattering produced
is mainly in one plane, and (ii) hemispherical diffusers
where the surface produces scattering in two or more coefficients due to finite panel size effects. This is why the
planes. (An insert in Figure 16 shows examples of the two flat surface generates diffuse reflections and consequently
types). For single plane diffusers the polar responses are large diffusion coefficients at low frequencies.
measured over a semicircle, for a hemispherical diffuser The diffusion coefficient definition is not compatible
the responses are measured over a hemisphere. This con- with most (if not all) geometric room acoustic models.
trasts with the ISO random incidence scattering coefficient There is no reason why a room model could not be devel-
which only gives a single hemispherical estimation even oped to use the diffusion coefficient definition, but philo-
for single plane diffusers (anisotropic surfaces). sophically the models are closer to the scattering coeffi-
The diffusion coefficient evaluates all mechanisms as cient which measures the amount of energy scattered away
diffuse reflection; the effects of surface roughness and from the specular direction. The scattering coefficient def-
edge diffraction are combined. AES-4id-2001 states that inition is not that useful for diffuser designers because
a reference flat surface should also be measured for com- it is concerned with how much energy is removed from
parison. This enables the determination of the point at the specular direction and not how the scattered sound is
which surface roughness becomes significant to be seen, distributed; as shown before, redirected energy can be in-
Figure 5b for example. One of the consequences of this, is terpreted as dispersion by scattering coefficients. In room
that at low frequency flat surfaces produce large diffusion models, the scattered sound is usually distributed accord-

8
Cox et al.: Scattering and diffusion coefficients tutorial ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 92 (2006)

source positions should be in the far field, otherwise the


1
sound pressure level measured includes near field aberra-
AES-4id minimum distance tions. This can be a problem with measurement, but is not
Diffusion coefficient

0.75 a problem with prediction, as receivers can be placed a


long distance away in that case. Figure 10 shows how the
0.5 diffusion coefficient varies with receiver distance from the
Rectangular battens
surface. The near field extends for a long distance from
0.25 Plane surface the surface, particularly for oblique sources and receivers.
Furthermore, equation (3) is monitoring the uniformity of
0
0.1 1 10 100
energy polar responses, and close to large surfaces specu-
Receiver arc radius (m) lar reflection effects dominate. If the receivers are too close
to the surface, they all receive similar energy from a flat
surface. Consequently, measurements must be made some
Figure 10. The variation of the diffusion coefficient with receiver way from the surface so that a majority of receivers do not
distance for two surfaces. The minimum measurement radius for
receiver specular reflections. This then enables the energy
this geometry given in AES-4id-2001 is marked (After Harg-
scattered away from specular directions to be evaluated.
reaves et al. [40]).
Fortunately, it is not necessary to measure at distances
where the scattering energy follows the inverse square law
ing to Lambert’s law rather than a uniform energy distribu- – the true far field – because that is a long way from the
tion, another incompatibility with the diffusion coefficient surface. The process of getting the diffusion coefficient is
definition. Lambert’s law is not used in diffuser design be- to first gain the polar response data, and then throw away
cause it is possible to achieve even energy distribution, a most of the information in data reduction. Consequently,
philosophy that can be traced back to Schroeder’s orig- being overly exacting in gaining the polar response and
inal work on diffusers. Furthermore, Lambert’s law is a insisting on measurement in the true far field is not nec-
high frequency, incoherent point scattering law, whereas essary. Incidentally, for some surfaces it is not sufficient
specialist diffusing surfaces deliberately exploit coherence just to measure in the far field. For concave surfaces, and
and interference to generate their diffuse reflections. Con- others that might have significant aberrations closer to the
sequently, Lambert’s law is inapplicable for evaluating dif- surface, it is necessary to measure in the near field as well
fusers. Conversely, using the diffusion coefficients in geo- as the far field to ensure that effects such as focussing are
metric room acoustic models is likely to produce incorrect monitored.
results. Further measurement problems arise for hemispherical
evaluation because the number of source and receiver posi-
4.2. Advantages tions to be measured becomes large if many angles of inci-
dence need be measured. For a single angle of incidence,
A key advantage of the diffusion coefficient formulation over a thousand receiver positions must be measured at
is that evaluation can be carried out using prediction a resolution of 5◦ . This requires an automated measure-
models. The polar response scattering from diffusers can ment system. To get random incidence values, assuming
be predicted accurately using techniques such as BEMs a 10◦ incident angle resolution, an impractical 0.4 million
[42, 43, 44] and so the diffusion coefficient can be pre- source-receiver pairs need to be measured. (The number of
dicted. This enables diffuser designs to be evaluated with- measurement points can be reduced in many cases because
out having to resort to the construction of prototypes and surfaces have planes of symmetry, but this would still lead
time consuming measurements. In the last decade, a dif- to an impractically large data set). Such detailed evalua-
fusion coefficient has been used in diffuser design, for ex- tion, however, is rarely needed and so AES-4id-2001 sug-
ample in optimisation. Anecdotal evidence suggests that gests a smaller number of incidence angles that can be
structures that have been developed using the diffusion co- used if time is limited. There are two main reasons why
efficient are successful [13]. Consequently, a track record very detailed random incidence is not required. (i) Acous-
of successful use of the diffusion coefficient in diffuser de- tics is governed by reciprocity, by measuring a large num-
sign and application is being built up. ber of source-receiver pairs there is a large amount of du-
plication. (ii) Diffusers are usually applied to treat first or-
4.3. Difficulties der reflection issues, and consequently the first order an-
A key difficulty with the diffusion coefficient is the near gles of incidence are the most important to measure and
field problem which hampers measurements of large sur- other angles of incidence are of secondary importance.
faces. The diffusion coefficient is easiest to measure when Single plane measurements and predictions do not suffer
surfaces are small, consequently, it is often said that the from this problem, and are quick and easy to carry out.
diffusion coefficient is for small surfaces with large rough- The main drawback in using the autocorrelation func-
ness. The problem with large surfaces arises because of tion to evaluate diffusion from the polar response is that
near field effects. To get results from equation (3) that intermediate values of the diffusion coefficient are hard to
are invariant to source and receiver radii, the receiver and interpret. For example, what does a value of 0.5 mean? It

9
ACTA ACUSTICA UNITED WITH ACUSTICA Cox et al.: Scattering and diffusion coefficients tutorial
Vol. 92 (2006)

be generated without temporal dispersion. This can lead to


1
echoes not being sufficiently suppressed [14].
Diffusion coefficient

0.8
5. Contrasting diffusion and scattering co-
0.6 efficient
0.4
Comparisons between diffusion and the scattering coeffi-
0.2 cients have rarely been undertaken; they have been devel-
oped in relative isolation. Comparisons are hindered be-
0
125 200 315 500 800 1.25k 2k 3.15k
cause by definition many surfaces that could be tested us-
f (Hz) ing the diffusion coefficient can not be tested using the
scattering coefficient and vice versa. The following sec-
tions detail some comparison work, which goes some way
Figure 11. The diffusion coefficient for various sets of semicylin- to illustrating and comparing the differences between scat-
ders. The number of cylinders in each set is: ——: 1; – – –: 2; tering and diffusion coefficients, although it has not always
— —: 3; : 4; - - - - -: 8. (After Cox and D’Antonio [14]).
been possible to achieve an exact comparison of the ISO
and AES methods.
is hoped that overtime an understanding will be built up
of what intermediate values mean and so this will only be 5.1. Schroeder diffusers and Fourier theory
a short lived problem. For example, experience indicates The far field scattering from a Schroeder diffuser can be
that the difference limen for the diffusion coefficient for predicted using a simple Fourier Model [44]. The Fourier
single plane diffusers is about 0.1, i.e. diffusers with coef- Model is not exact, but it does give reasonably accurate
ficients within 0.1 of each other have equally good polar predictions of the scattering from the surfaces except at
responses. low frequencies and large angles of incidence or reflection.
Under this approximate model, the scattered pressure, ps ,
4.4. Other features from the surface can be simplified to:
The diffusion coefficient values produced tend to be small.
 
Np N
Values for the autocorrelation coefficient can in theory ps (ψ, θ) = A e−2jkdn e jk(n+np N)w(sin θ+sin ψ), (4)
spread over the entire range from 0 to 1. A value of close to np =1 n=1
zero has been measured for a concave surface designed to
focus sound on a single receiver. A value of 1 can be mea- where ψ is the angle of incidence, θ the angle of reflection,
sured for a small single sphere or semi-cylinder. As soon Np the number of periods, N the number of wells in a pe-
as more complex surfaces are introduced such as a set of riod, w the well width, k the wavenumber, dn the depth of
semi-cylinders, the diffusion coefficient reduces because the nth well, and A is a constant. This approximate theory
of the lobing introduced. This is illustrated in Figure 11. enables a simple formulation for a scattering coefficient to
This lobing is unavoidable in extended structures, and so be derived. In addition, the polar response can also be cal-
the diffusion coefficient is rarely close to 1 for usable culated from equation (4) and so the diffusion coefficient
and realistic diffusers. For example, a single semi-cylinder can be found from equation (3). Hence, a direct compar-
may produce complete diffusion, but to cover a wall a set ison of diffusion and scattering coefficients can be made
of semi-cylinders are needed, and the mutual interference [45]. The scattering coefficient is not the ISO random in-
between the reflections from each of the cylinders mean cidence coefficient, however, but instead is a free field ver-
that diffusion is no longer complete. This phenomena is sion of the Mommertz and Vorländer coefficient [2, 3].
why AES-4id-2001 makes detailed requirements on the The free-field scattering coefficient is evaluated by find-
test surface, for example insisting that if a surface is to ing the invariant energy, Espec , in the specular direction
be applied in a periodic manner, it is necessary to measure (ψ = −θ) when the surface is moved. Equation (4) is a
at least four repeat sequences of the surface. Experience single plane formulation, so it is natural to translate the
shows that for practical surfaces, the diffusion coefficient surface. If the surface is assumed to be very large, so that
does not exceed 0.7. edge effects are not significant, then the scattered pressure
While a diffusion coefficient is a useful measure of spa- does not change with translation. The invariant energy is
tial dispersion, it does not monitor the degree of temporal found by averaging over the translation of one period:
dispersion generated, i.e. the amount that the reflections  2
are dispersed in time in the impulse response. If a diffuser  Np N 
    −2jkd 
is being used to treat a first order reflection problem such Espec 
≈ A e n
(5)
 ,
as an echo, the degree of temporal dispersion is impor-  np =1 n=1 
tant to the suppression of the echo. For complex surfaces,
spatial and temporal dispersion go hand-in-hand, but for where A is a constant. The invariant energy has to be put
simple surfaces, such as cylinders, spatial dispersion may in a ratio with the energy from a flat plane surface for nor-

10
Cox et al.: Scattering and diffusion coefficients tutorial ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 92 (2006)

Scattering / Diffusion coefficient


1

0.8

0.6

0.4

0.2

0
0 1000 2000 3000 4000
f (Hz)

Figure 12. Diffusion and correlation scattering coefficients for a Figure 13. Scattered polar distribution for - - - - -: N = 7 QRD,
modified primitive root diffusing using the Fourier test bed. The and ——: N = 7 QRD with the well order rearranged. f =
design frequency was 500 Hz, and the diffuser has 6 wells per pe- 4500 Hz, QRD design frequency 500 Hz. Simple Fourier theory.
riod. – – –: Scattering coefficient using Equation (7); : Dif-
fusion coefficient, diffuser; · · · · · ·: Diffusion coefficient, plane
surface; ——: Correlation scattering coefficient. (Modified from - there is no dependence on the order of the wells in the dif-
Cox and D’Antonio [45]). fuser. Figure 13 shows the polar responses for a QRD and
a QRD with the well order rearranged. The QRD has more
malisation purposes [3]. This ratio gives a reflection coef- even lobe energy than the reordered QRD and so the AES
ficient, Rspec : diffusion coefficients are different, 0.19 and 0.10 respec-
tively. The free field scattering coefficient, however, re-
  
  Np N −2jkdn 2 mains unchanged because in both cases the energy moved
   A np =1 n=1 e  from the specular direction is the same. This is an illus-
Rspec 2 =  2 . (6)
A Np N  tration of why the diffusion coefficient is seen as a stricter
test of diffuser quality.
This then represents the proportion of energy that is spec- Sakuma and Kosaka [22] examined different types of
ularly reflected by the surface, and so the scattering coeffi- scattering coefficient using BEM predictions on sinusoidal
cient can be readily evaluated. To emphasis that this is not surfaces. They refined the Mommertz and Vorländer free-
the same as the ISO random incidence coefficient, it will field method by using more receiver points. Instead of just
be given the symbol s : using a receiver located in the specular reflection, Sakuma
 2 and Kosaka used many receivers placed over the whole
1 
N  hemisphere. The energy variant and invariant to surface
 
s = 1 −  e−2jkdn  . (7) movements is calculated at each receiver, and these then
N 
n=1 averaged and used to give the scattering coefficient. They
The scattering coefficient is independent of angle of inci- demonstrated that the coefficient evaluated this way bet-
dence, a peculiarity for the surface being tested here. Con- ter matches the random incidence ISO coefficient than the
sequently, the random incidence scattering coefficient is original Mommertz and Vorländer free-field method for
numerically identical to the free-field case. (Incidentally, surfaces where the scattering is strongly angular depen-
equation (7) is similar to the prediction model Embrechts dent. (For the Schroeder diffusers being examined here,
et al produced for Gaussian rough surfaces - see equation the scattering is not sufficiently angular dependent to show
(19) in reference [46] - except for a cos() factor in the ex- the difference between the evaluation techniques).
ponent.)
Equation (7) shows that to get the greatest scattering, 5.2. Correlation scattering coefficient
the sum of the reflection coefficients of the wells must
Mommertz presented a method for evaluating a free field
be evenly spaced around the unit circle. This is achieved
scattering coefficient from polar responses of single plane
for the modified versions of the primitive root diffuser
diffusers measured in the plane perpendicular to the corru-
[47, 48], at integer multiples of a design frequency. This
gations. This is done by correlating the scattered pressure
is illustrated in Figure 12 where at multiples of the design
polar responses from the test surface and a reference flat
frequency, s = 1 (except at 3 kHz where it behaves like
surface. This correlation scattering coefficient is given by:
a flat surface). This complete scattering simply means no
energy is in the specular direction, it doesn’t necessarily  n 2
 ∗ 
say how good the dispersion produced is. This is why the i=1 p1 (θi )p0 (θi )
δc = 1 −   2  n  2 , (8)
AES coefficients are numerical less than the scattering co- n    
i=1 p1 (θi ) i=1 p0 (θi )
efficient.
This free field scattering coefficient in Equation (7) is where p1 is the pressure scattered from the test surface;
given by the sum of the well reflection coefficients squared p0 is the pressure scattered from the flat surface, and θi

11
ACTA ACUSTICA UNITED WITH ACUSTICA Cox et al.: Scattering and diffusion coefficients tutorial
Vol. 92 (2006)

the receiver angle of the ith measurement position. An al-


ternative description of this coefficient was given by Em- 1
brechts et al. [46] who described it as an LMS prob-

Scattering coefficient
0.8
lem and extended the definition for the more general case
of hemispherical diffusers. Cox and D’Antonio [14] have 0.6
produced extensive tables of this scattering coefficient for
single plane scatterers. In general, scattering coefficients 0.4
are needed in 1/3 octave bands, but the definition given in 0.2
Equation (8) is necessarily a single frequency definition
because pressure magnitude and phase is required. There 0
0 1000 2000 3000 4000
are various methods that can be devised for getting a 1/3
f (Hz)
octave band value, but they all give numerically similar
values. Consequently, it was found that a simple average
Figure 14. Predicted and measured correlation scattering coef-
of the single frequency scattering coefficient values pro- ficient. 2D measurements on semi-cylinders compared to BEM
duced in each 1/3 octave band is sufficient. predictions. ——: Prediction, 1 cylinder; : Measurement, 1
The correlation scattering coefficient is not the same cylinder; – – –: Prediction, 4 cylinders; - - - - -: Measurement, 4
as the ISO random incidence coefficient or the free-field cylinders. (After Cox and D’Antonio [14]).
scattering coefficient calculated using surface movements.
This is illustrated in Figure 12 where the free-field scatter-
ing coefficient using surface movement is compared to the
correlation scattering coefficient for the simple Schroeder 1.2
diffuser test bed. This difference arises because the coeffi- Scattering or Diffusion coefficient
cient definition is different. The free-field Mommertz and 1
Vorländer method measures the amount of energy moved
0.8
from the specular direction when the surface is moved, the
correlation scattering coefficient measures the dissimilar- 0.6
ity between the test and flat surface scattering over a polar
response. 0.4

One useful property of the correlation scattering coef- 0.2


ficient is that it is readily predicted; something that is not
0
true of the ISO random incidence scattering coefficient.
100 500 1000 4000
Kosaka and Sakuma [49] examined various issues con- f (Hz)
cerning the calculation of free-field scattering coefficients
using BEM models. They concluded that the minimum re-
Figure 15. Various scattering and diffusion coefficients for a
ceiver distance was the sample diameter, that the receiver model scale N=7 QRD shown in Figure 16. ——: Correlation
points should be spaced at most 3◦ apart, and it does not scattering coefficient averaged over 3 sample orientations and
matter for the correlation scattering coefficient whether the 3 sources; : Random incidence ISO scattering coefficient;
sample is round or square. However, they also found that - - - - -: Single plane diffusion coefficient, diffuser; – – –: Single
edge diffraction can cause the scattering coefficient to be plane diffusion coefficient, plane surface. —×— Single plane scat-
overestimated, something that can also be seen in the scat- tering coefficient predicted using equation (7).
tering coefficient tables in [14].

5.3. Some results and discussions the wells parallel to the measurement arc. For each sam-
ple orientation, three incidence angles (0◦ , 30◦ and 60◦ )
Here predictions have been tested by comparing predic- were measured. These measurement results were then av-
tion and measurement in a two dimensional polar response eraged to give an approximate random-incidence coeffi-
for a single cylinder and a set of cylinders. Predictions cient. A similar sample was measured using the ISO ran-
were carried out using a BEM [42] and measurements in dom incidence method in a model reverberation chamber.
a 2D goniometer [36]. Figure 14 compares the predicted The results are compared in Figure 15. There is reasonable
and measured correlation scattering coefficient and a good correspondence between the two measured results, though
match is achieved. This provides evidence that the coef- in two frequency bands the results are significantly differ-
ficient can be predicted and that the measurement system ent. For example, for the 3.2 kHz octave band the random
used is robust. incidence measurements exceeds 1, something that can not
To measure the random incidence correlation scattering happen with the correlation scattering coefficient. Consid-
coefficient, a sample of a 1D QRD has been tested at 1:5 ering one measurement is done in a diffuse field and the
scale. It is shown as an insert in Figure 16. The sample other in the free field, however, the match is actually quite
was measured in the 2D goniometer first with the QRD good, indeed better than many have obtained when diffuse
wells perpendicular to the measurement arc, and then with and free field absorption coefficients are compared.

12
Cox et al.: Scattering and diffusion coefficients tutorial ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 92 (2006)

Also shown in Figure 15 is the scattering coefficient pre-


dicted using the simplest Fourier model, Equation (7). The 1.2
prediction accuracy is surprisingly good considering that
1

Scattering coefficient
the Fourier theory makes many assumptions. Figure 15
also shows the diffusion coefficient. This is a single plane 0.8

measurement in the plane of maximum diffusion. In this 0.6


case the scattering and diffusion coefficients agree as to the 0.4
frequency at which significant scattering/diffusion begins
0.2
(≈500 Hz). Again, the diffusion coefficient is numerically
0
less than the scattering coefficient values. 800 1260 2000 3125
125 200 320 500
A final feature of note in Figure 15, is that the 1D QRD Frequency (Hz)
produces a high value for both the ISO and the correlation
scattering coefficients, even though it is plane and extruded
in one direction. Figure 16 further illustrates this point,
where the ISO random incidence scattering coefficients for
a single plane and a hemispherical diffuser are compared.
To use a very simplistic analysis for the 1D QRD, even
if the scattering coefficient in the plane of maximum dis-
persion was 1, the scattering coefficient in the extruded
direction must be close to 0, and so it might be expected
that the hemispherical coefficient would be ≈0.5. Yet high
Figure 16. Random incidence ISO scattering coefficients for two
frequency values for the random incidence scattering co- different diffusers and also the diffuser shapes. Left diffuser is
efficient are close to 1; this happens because when the sur- single plane device, right diffuser is hemispherical device. Mul-
face is rotated, the topology changes dramatically. Conse- tiple periods of each were used. 8 periods of the single plane
quently, the ISO method may often produce high scatter- device equalled the diameter of the base plate. : Single plane,
ing coefficients for extruded 1D surfaces. A more strict 2 : Hemispherical. (After Cox and D’Antonio [14]).
measure of scattering ability would be two coefficients
in two orthogonal directions, as is done for the diffusion
coefficient. But then current geometric room models can While the scattering coefficient has been developed for
only deal with single hemispherical based scattering coef- room acoustic modelling, there has been insufficient test-
ficients, so this more strict evaluation is incompatible with ing within the room models. While some have found good
current room models. Some models are introducing sev- accuracy using the values, other have not. The discrep-
eral distribution functions to be used in conjunction with ancies appear to centre on the type of room being mea-
the scattering coefficient; this may help in more accurately sured. If the overall room shape and sizes and orientations
distributing the randomly directed diffuse rays. of surfaces are such that a diffuse field is created even if
no rough or scattering surfaces are used, then the reverber-
ation time can be well predicted even without diffuse re-
6. Discussions and conclusions
flection modelling, even if predictions of finer parameters
In order to fully understand and model the physical prop- such as clarity may suffer [50]. For other room shapes, ac-
agation of sound in a space, the wave nature of sound, and curate modelling of scattering is more crucial. As the value
the diffusion and scattering that finite and rough surfaces of the scattering coefficient required depends on the room
produce, must be considered. It is necessary to include a model being used [9], blindly applying measured scatter-
model of surface scattering in geometric models to enable ing coefficients in room models may produce inaccurate
accurate predictions in many cases. This surface scattering predictions for markedly non-diffuse spaces.
model requires a coefficient that determines the proportion Scattering coefficients give a quick and rough estimate
of energy specularly and diffusely reflected, and to date of the scattering process, they should not be used to eval-
this has been done by precedent and empirically. With the uate the worth of surfaces when designing or specifying
standardisation by ISO of a method to measure the scatter- diffusers. The scattering coefficient is only concerned with
ing coefficient, it should be possible to refine and improve how much energy is moved from the specular direction, it
room predictions. In most practical cases users of geomet- does not measure the quality of dispersion. For this reason,
ric room acoustic models still have to guess most values, diffusing surfaces need to be evaluated using the diffusion
but if a set of typical wall elements in concert halls (say coefficient when quality is being assessed. The diffusion
coffered ceilings of various depths) are eventually mea- coefficient should not, however, be blindly used in geomet-
sured, they can serve as guidelines for the actual ones - ric room models as its definition is not compatible with the
much like absorption coefficients today where an exact surface scattering models used in current geometric algo-
measured case is seldom found (except for commercial ab- rithms.
sorbers) but some “engineering skills” have to be applied There are many issues surrounding these coefficients
to estimate a value based on similar measured ones. that remain to be resolved. One common question is

13
ACTA ACUSTICA UNITED WITH ACUSTICA Cox et al.: Scattering and diffusion coefficients tutorial
Vol. 92 (2006)

whether there is a direct link between the diffusion or scat- [14] T. J. Cox, P. D’Antonio: Acoustic absorbers and diffusers:
tering coefficient and a physical property of the space. This Theory, design and application. Spon Press, 2004.
arises because practitioners are used to the direct link be- [15] D. Davis, C. Davis: The LEDE concept for the control of
tween the absorption coefficient and the reverberation time acoustic and psychoacoustic parameters in recording con-
trol rooms. J. Audio Eng. Soc. 28 (1980) 585–95.
that arises in a diffuse field. For diffuse reflection and scat-
tering there is no simple relationship, but maybe future re- [16] P. D’Antonio, J. Konnert: The RFZ/RPG approach to con-
trol room monitoring. Proc. Audio Eng. Soc., October
search should include investigating what relationships, if 1984, Preprint 2157 (I–6).
any, exist between these coefficients and the room acous-
[17] T. J. Cox: Acoustic diffusers: the good, the bad and the ugly.
tic quality. Proc. IoA(UK) 26 (Reproduced Sound (2004)).
Another avenue for future research is to develop meth-
[18] T. J. Cox: Optimization of profiled diffusers. J. Acoust. Soc.
ods for in-situ measurement of the diffusion or scattering Am. 97 (1995) 2928–41.
coefficients. There is a great deal of interest in measuring
[19] T. J. Cox: Designing curved diffusers for performance
in-situ absorption coefficients [51]. In recent round robin spaces. J. Audio Eng. Soc. 44 (1996) 354–364.
trials of room acoustic models [52], a key finding has been [20] J. J. Embrechts: Broad spectrum diffusion model for room
that the accuracy of the prediction models are highly de- acoustics ray-tracing algorithms. J. Acoust. Soc. Am. 107
pendent on the quality of the input data, including the ab- (2000) 2068–81.
sorption and scattering coefficients of the surfaces. One [21] R. Heinz: On modelling the diffuse scattering properties of
of the difficulties in carrying out an in-situ measurement boundaries in sound particle simulations for room acous-
of scattering is that the effects of absorption and diffusion tics. Acustica 82 (1996) 82–90.
can be similar on the volume acoustic, making it difficult [22] T. Sakuma, Y. Kosaka: Comparison between scattering co-
to separate the two effects. efficients determined by specimen rotation and by directiv-
ity correlation. International Symposium on Room Acous-
tics: Design and Science (RADS), Awaji, Japan, 2004, 056.
References [23] E. Mommertz: Determination of scattering coefficients
from reflection directivity of architectural surfaces. Ap-
[1] ISO/FDIS 17497-1: Acoustics - Measurement of the sound
plied Acoustics 60 (2000) 201–203.
scattering properties of surfaces – Part 1: Measurement of
the random-incidence scattering coefficient in a reverbera- [24] A. Farina: A new method for measuring the scattering co-
tion room. 2000. efficient and the diffusion coefficient of panels. Acustica 86
(2000) 928–942.
[2] E. Mommertz, M. Vorländer: Measurement of scattering
coefficients of surfaces in the reverberation chamber and [25] T. J. Hargreaves: Acoustic diffusion and scattering coeffi-
in the free field. Proc. 15th ICA,, 1995, II, 577–580. cients for room surfaces. Ph.D. Thesis, University of Sal-
ford, 2000.
[3] M. Vorländer, E. Mommertz: Definition and measure-
ment of random-incidence scattering coefficients. Applied [26] ISO 354: Acoustics - Measurement of sound absorption in
Acoustics 60 (2000) 187–199. a reverberation room. 2003.
[4] AES-4id-2001: AES information document for room acou- [27] J. J. Embrechts: Practical aspects of the ISO procedure for
stics and sound reinforcement systems – characterisation measuring the scattering coefficient in a real-scale exper-
and measurement of surface scattering uniformity. J. Audio iment. Proc. Forum Acusticum, Sevilla, 2002, RBA–06–
Eng. Soc. 49 (2001) 149–165. 001–IP.
[5] H. Kuttruff, T. Strassen: On the dependence of reverbera- [28] L. De Geetere, G. Vermeir: Investigations on real-scale ex-
tion time on the wall diffusion and on room shape. Acustica periments for the measurement of the ISO scattering coef-
45 (1980) 246–55. ficient in the reverberation room. Proc. Forum Acusticum,
Sevilla, 2002, RBA–06–004.
[6] Y. W. Lam: On the parameters controlling diffusion calcu-
lation in a hybrid computer model for room acoustic pre- [29] M. Vorländer, J. J. Embrechts, L. De Geetere, G. Vermeir,
diction. Proc. IoA 16 (1994) 537–544. M. H. D. Gomes: Case studies in measurement of random
incidence scattering coefficients. Acta Acust. united with
[7] M. Hodgson: Evidence of diffuse surface reflections in Acust. 90 (2004) 858–67.
rooms. J. Acoust. Soc. Am. 89 (1991) 765–71.
[30] M. Vorländer, H. Bietz: Der Einfluß von Zeitvarianzen
[8] M. Vorländer: International round robin on room acoustical bei Maximalfolgenmessungen. Fortschritte der Akustik,
computer simulations. Proc. 15th ICA, 1995, II, 689–692. DAGA 95, 1995, 675–678.
[9] Y. W. Lam: A comparison of three diffuse reflection mod- [31] M. Vorländer, M. Kob: Practical aspects of MLS measure-
eling methods used in room acoustics computer models. J. ments in building acoustics. Applied Acoustics 52 (1997)
Acoust. Soc. Am. 100 (1996) 2181–92. 239–258.
[10] B.-I. L. Dalenbäck, M. Kleiner, P. Svensson: A macro- [32] G. B. Stan, J. J. Embrechts, D. Archambeau: Comparison
scopic view of diffuse reflection. J. Audio Eng. Soc. 42 of different impulse response measurement techniques. J.
(1994) 793–807. Audio Eng. Soc. 50 (2002) 249–62.
[11] H. Kuttruff: Room acoustics, 4th edition. E. & F. N. Spon, [33] A. Farina: Simultaneous measurement of impulse response
2000. and distortion with a swept-sine technique. Proc. Audio
[12] M. R. Schroeder: Binaural dissimilarity and optimum ceil- Eng. Soc., 108th convention, 2000, preprint 5093.
ings for concert halls: More lateral sound diffusion. J. [34] M. H. A. Gomes, M. Vorländer, S. N. Y. Gerges: Aspects
Acoust. Soc. Am. 65 (1979) 958–63. of the sample geometry in the measurement of the random-
[13] P. D’Antonio, T. J. Cox: Diffusor application in rooms. Ap- incidence scattering coefficient. Proc. Forum Acusticum,
plied Acoustics 60 (2000) 113–42. Sevilla, 2002, RBA–06–002–IP.

14
Cox et al.: Scattering and diffusion coefficients tutorial ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 92 (2006)

[35] T. J. Cox, P. D’Antonio: Surface characterization for room [44] T. J. Cox, Y. W. Lam: Prediction and evaluation of the scat-
acoustic modelling and design. Proc. International Sym- tering from quadratic residue diffusers. J. Acoust. Soc. Am.
posium on Room Acoustics: Design and Science (RADS), 95 (1994) 297–305.
2004, CO2. [45] T. J. Cox, P. D’Antonio: Contrasting surface diffusion and
[36] P. D’Antonio: The Disc Project: Directional scattering co- scattering coefficients. Proc. 17th ICA, 2001, 6B.09.01.
efficient determination and auralization of virtual environ- [46] J. J. Embrechts, D. Archambeau, G. B. Stan: Determina-
ments. Proc. Noise-Con 93, May 1993, 259–264. tion of the scattering coefficient of random rough diffusing
[37] D. Takahashi: Development of optimum acoustic diffusers. surfaces for room acoustics applications. Acta Acustica–
J. Acoust. Soc. Jpn. 16 (1995) 51–58. Acustica 87 (2001) 482–94.
[38] J. A. S. Angus, A. C. Marvin, J. Clegg, J. F. Dawson: A [47] E. Feldman: A reflection grating that nullifies the specular
practical metric for evaluating sound diffusers. Proc Audio reflection: A cone of silence. J. Acoust. Soc. Am. 98 (1995)
Eng. Soc., 98th Convention, February 1995, Preprint 3955 623–634.
(D5). [48] T. J. Cox, P. D’Antonio: Acoustic phase gratings for re-
[39] J. A. S. Angus: Diffuser assessment using surface spherical duced specular reflection. Applied Acoustics 60 (2000)
harmonics. J. Acoust. Soc. Am. 104 (1998) 1857. 167–186.
[40] T. J. Hargreaves, T. J. Cox, Y. W. Lam, P. D’Antonio: Sur- [49] Y. Kosaka, T. Sakuma: Numerical study on the behavior
face diffusion coefficients for room acoustics: free field of scattering coefficients of wall surfaces. Proc. 18th ICA,
measures. J. Acoust. Soc. Am. 108 (2000) 1710–20. 2004, Mo2.B2.2.
[41] P. D’Antonio, J. H. Konnert, P. Kovitz: The Disc Project: [50] B.-I. Dalenbäck: The importance of diffuse reflection in
Experimental measurement of the directional scattering computerized room acoustic prediction and auralization.
properties of architectural acoustic surfaces. Proceedings Proc. IoA(UK) 17 (1995) 24–34.
of Sabine Centennial, June 1994, 1pAAd2, 141–4. [51] C. Nocke, V. Mellert: Brief review on in situ measurement
[42] T. J. Cox: Predicting the scattering from reflectors and dif- techniques of impedance or absorption. Forum Acusticum,
fusers using 2-dimensional boundary element methods. J. Sevilla, 2002, RBA–03–001–IP.
Acoust. Soc. Am. 96 (1994) 874–8. [52] I. Bork: A comparison of room simulation software – the
[43] T. J. Cox, Y. W. Lam: Evaluation of methods for predicting 2nd round robin on room acoustical computer simulation.
the scattering from simple rigid panels. Applied Acoustics Acustica 86 (2000) 943–956.
40 (1993) 123–140.

15

View publication stats

You might also like