The optimization of profiled diffusers

Trevor J. Cox
Schoolof ESD, SouthBank University,103 BoroughRoad,LondonSE10AA, UnitedKingdom

(Received11August1993;revised12 December1994;accepted
20 December1994)
Methodshavebeendevelopedto produceprofileddiffusersthatcreatea largeamountof diffusion.
The methodsare iterativeandrequiredthe development of a new parameterto measurediffusion.
Achievingscatteringindependent of anglehasbeenattemptedovera wide bandwidth.The methods
are alsoapplicableto otherdiffusioncriteria.The diffusersconsistsof a seriesof wells of the same
width but of differentdepthssimilar to Schroederdiffusers.Applicationsincludeconcerthalls,
theatres,and studio monitor rooms. The new diffusershave been shown to create better, more
uniformdiffusionthan the previousdesignsof Schroeder.This is due to the new designsbeing
relianton accurateboundaryelementpredictionmethodsratherthanmoreapproximate techniques.
PACS numbers: 43.55.Br, 43.20.Fn

INTRODUCTION ers has been made oncebefore.5 This was restrictedto nar-
row deep wells and a single frequency.The narrow deep
Profiled diffusersbased on wells were introducedby
wells would not necessarilybe appropriatefor usein audito-
Schroeder;
• an example
of oneis shown
in Fig. l(a). The
one-dimensional form of these diffusers consists of a series
ria whereminimizing'absorption
is essential,
andwideband
diffusion is required.Furthermore,the optimizationpro-
of wells, of the samewidth,but of differentdepthsseparated
cessesand evaluationof the scatteringrelied on a relatively
by thin fins.The diffusersare long anduniformin onedirec-
simplepredictiontheory.In this paper all the resultingscat-
tion sothe scatteringis mainlyin oneplanedominatedby the
tering is evaluatedusing an accurateboundaryelement
effectsof the well depthsequence.
Schroeder diffusersarebasedon thefollowingconcepts.
method.
4 Thishasalsobeenusedfortheoptimization
pro-
cesseswhen computationtime was not excessivelylong.
When soundis incidenton the diffusersplanewavespropa-
gateup anddown eachwell. The wavesthenradiatefrom the
wells and createan interferencepattern.The relativephases I. PREDICTION METHODS
of theseradiatingwavescanbe alteredby changingthe well
depths.By choosingan appropriatewell depthsequence,a
To predictthe scatteringfrom the diffusers,the fact that
particulardesiredinterferencepatterncan be obtained.To
they are uniform and long in one directionwas exploited.
choosea suitabledepth sequenceSchroederexploitedthe
This allowedtwo-dimensional predictionmethodsto be ap-
fact that the Fouriertransformof a "surface"approximately
plied to a crosssectionthroughthe diffuser.Thesemethods
givesthe far-fielddiffractedpressuredistribution.
If the qua-
producegood predictionsof the diffractedpressure distribu-
draticresiduesequence is usedto determinethe well depths,
the Fourier transform of the surface is constant. Hence such tion,butnot'•he
overall
magnitude.
6Thisisnotaproblem
for
this investigationwhere the scatteredpressuredistribution
a diffusershouldproduceoptimumdiffusion.Due to ap-
only is required.
proximationsnecessaryin the designtheory,however,the
Prediction methodsbased on two boundaryelement
true scatteringis only ever an approximationto optimum
methods(BEMs) formulatedfrom the Helmholtz-Kirchhoff
diffusion. More detailed discussionsof the design of
Schroeder diffusers can be found in Refs. 1-4. integral equationhave been used.These were the standard
In previous
workondiffusers
usedin auditoria,
4 a vari- BEM and thin panel BEM. Brief details are given below.
ety of methodsfor predictingthe scatteringfrom Schroeder This is followed by a descriptionof how the methodsare
diffusershavebeendevelopedand tested.As shallbe shown appliedto the variousdiffusers.
later, it is possibleto usethesemethodsin an iterativeman-
ner to producediffuserswhich are betterthanthosebasedon A. Helmholtz-Kirchhoff integral equation
the quadraticresiduesequence.
The predictionmethodswere based on the single fre-
It is also possibleto apply this iterativemethodto dif-
fusers of different construction to Schroeder diffusers. This quencyform of the Helmholtz-Kirchhoff integralequation
hasbeendemonstrated
by optimizingone othertype of dif- and its normal derivative.For a single sourceat r 0 and a
fuser which can be described as a Schroeder diffuser without surface
S, theygivethepressure
P(r) anditsderivative
as7
fins. This type of surfacewill be referredto as a stepped
diffuser;an exampleis shownin Fig. l(b). Being able to
constructdiffuserswithoutthesefins is an advantageto ac-
ousticians. These fins add to construction costs and are one
of the mostlikely sourcesof absorption.
aP(r)=•s(•G(r'rs) ikfi(rs)G(r, rs) e(r,)as

An investigationinto the optimizationof profileddiffus- +Pi(r, ro), (1)

2928 J. Acoust.Soc.Am. 97 (5), Pt. 1, May 1995 0001-4966/95/97(5)/2928/9/$6.00 ¸ 1995 AcousticalSocietyof America 2928

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 146.189.194.69 On: Fri, 19 Dec 2014 23:56:07
 3D view Where possiblesymmetrywas exploited.In that case only
half the diffuser is modeled and the principle of image
sourcesused;this decreasespredictiontimes.From this, the
Green'sfunctionis alteredto includeimage sources:

(a) (b) G(r,rs)=-(i/4)[H(o1)(klr-rsl)+H(o1)(klr-r•l)],

(4)
where
r• istheposition
o•theimage source.
The solutionof the Helmholtz-Kirchhoff integralequa-
Cross sections tion in the thin panellimi,twas alsoused.Considerthe spe-
cial caseof a rigid panel whosethicknessreducesto be in-
fin finitely thin. Figure 2 shall be usedas an exampleof sucha
body.The following nomenclatureis used:a subscripti or 2
is introducedfor the front and back surfaces,and n(rs•) is
writtenasn. It canthenbe shown
7 thatEqs.(1) and(2)
become
(a) (b)

0: L[P(rs•)-P(rs2)]
32G(r,
rs
) dS
FIG. 1. Thediffuserstested:(a) Schroeder-type
diffuser,(b) stepped
diffuser 1 On(r)
On(rs)
without fins.

3Pi(r, r0)
(5)
o•P(r) oCG(r,rs) On(r)
On(r)
=isOn(r)n(rs) 0G(r, rs) '
P(r)=
L[P(rsl)-P(rs2)]
an(rs)
1
dS + P i( r, r0),

3G(r,
rs)) 3Pi(r,
r0)
-ik/3(rs)On(r)P(rs)dS+
On(r)'
whereEq. (5) is valid onlywhenr lieson thesurface;Eq. (6)
(6)

(2) is for externalreceiverpositions.The surfaceintegralis now

whereP i(r,r0) is the soundpressure directfrom the source, only carriedout on one sideof the thin panelas the equations
G(r,rs) is the appropriate Green'sfunction,/5(rs)is the lo- are in termsof the pressuredifferenceacrossthe panels.
callyreactingsurfaceadmittance, andn(rs) is theunitvector Equations(1)-(6) form the basisof the boundaryele-
normalto the surface,pointingout of the surface.a can have ment methods.A BEM solutionfor the pressureat r can be
valuesof 0, 1/2, or 1 depending,respectively,on whetherthe found for a surfaceof known geometryand known local
point r lies within, on the surfaceof, or externalto the re- reacting
surface
admittance.
7Thegeneral
solution
method
is
flecting object. Figure 2 shows definitionsof the vectors to break the diffuser down into a set of surface elements.
used. Acrosseachelementit is assumedthat the pressureand ad-
The Green's function was taken to be the standard two- mittanceare constant.A setof simultaneous equationsis then
dimensionalform for free space: set up and solvedto obtain the surfacepressureon each of
theseelements.Oncethesurfacepressures areknownEq. (1)
G(r,rs)- -(i/4 )H(01)(kl
r- rsl), (3) or (6) is appliedto obtainthe externalpointpressure.
where
H[l)(x) is theHankelfunctionof order0. Thiswas
evaluated
usinga standard
polynomial
approximation.
8 B. Predicting the scattering from diffusers with fins
For diffusers with fins such as the Schroeder diffusers,
source two methods4 have been used.The first is basedon the thin
panel BEM, the secondon the standardBEM.

•, y receiver 1. Thin panel BEM

Whenthediffuseris brokendowninto a setof thinpanel
elements,all surfacescanbe modeledincludingthe thin fins.
The methodusesEqs.(5) and(6). The completeenclosure
of
the diffuserwith a zero velocity boundaryconditionensures
that the pressureson the rear of the nonfin elementsand
within the diffuser'sbody are zero. Obtainingthe external
point pressuresfrom these integral equationsis an estab-
point on lished
technique.
9Therewasa possibility
ofnonunique
orno
surface
X
n •/ /S1 solutionsat critical frequencies.
F o,gin I To checkfor uniquesolutions,the predictionsusingthis
methodwere comparedto thoseproducedby the standard
BEM describedbelow.As the thin panel BEM relies on the
FIG. 2. Geometryusedin predictionmethods. derivative of the Helmholtz-Kirchhoff integral equation,

2929 d. Acoust. Soc. Am., Vol. 97, No. 5, Pt. 1, May 1995 Trevor d. Cox: The optimizationof profileddiffusers 2929

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 146.189.194.69 On: Fri, 19 Dec 2014 23:56:07
 whereasthestandard
BEM relie}on thenormalHelmholtz- Forboth(i) and(ii) theCHIEFn method
wasavailable
Kirchhoff integral equation,it is highly unlikely that these to ensureunique solutions.This was not used during the
methods
wouldhavethesamecriticalfrequencie•s.
If there- optimizationprocess,only duringthe latter evaluationof the
sults were similar it was assumed that the solutions were scattering.
accurateand unique.
The thin panelBEM hasthe capabilityto produceaccu-
rateresults
evenwhe
n plane-•ave
propagation.
withinthe II. LIMITATIONS OF SCHROEDER'S DESIGN THEORY
wells breaksdown.(Thenmethodshaveto be employedto
ens•[e
unique
solutions,
1øasthestandard
BEMbecomes
in- Schroeder'soriginal theory can be derived from the
accurateandso cannot be usedfor comparison.) Prediction Helmholtz-Kirchhoffintegralequationby a methodanalo-
times can becomeprohibitivelylong for large complicated gousto Fraunhoferdiffractionin optics.First, Kirchhoff's
surfaces.
4 boundary
conditions
9 haveto be applied
to Eq.(1).These
boundaryconditionsgive valuesfor the surfacepressureon
2. Standard BEM thefrontof thediffuser
as[1+R(rs)]Pi(rs),whereR(rs)is
the reflectionfactor and Pi(rs) the incidentpressure.The
It is notpossibleto handlethethinfinsin thenumerical pressureson the back andsidesare neglected.This leadsto a
solutionusing the standardBEM, and so an approximate scatteredpressureP s(r) of
representationof thediffuseris required.The diffuseris rep-
resentedby a box with a variableadmittanceon the front
face. These front face admittancesare calculatedassuming es(r)=A
•S1
[1 +R(rs)]G(rs,ro)

thatthe planewavespropagateup anddownthewells induc-

3G(r, rs)
ing a phasechangewith no absorption.For this to be true,
hard surfaces,local reaction, and small radiation admittance x 3n(rs-••--ikfi(rs)G(r,
rs)dS. (7)
are assumed. Suchan approximate representationis a com- In termsof well depths,the admittanceand reflectionfactor
mon first stepwhen predictingthe scatteringfrom thesedif- are given by
fusers.
Once the diffuserhas been representedby this simpler 1-fi(rs)
geometry,it is then possibleto solve the systemusing a fi(rs)
=-itan[kd(rs)],
R(rs)= l+fi(rs)' (8)
standardBEM basedon Eq. (1). The CHIEF methodcanbe whered(rs) is the well depthat rs. An assumption
will be
usedto confirm
unique
solutions.
n Duringtheoptimization made that the sourceand receiverdistancesare large com-
process,however,the CHIEF methodwas n6glectedto de- pared to wavelength.Then the large argumentasymptotic
creasecomputationtime. expansion
fortheHankelfunction
canbeused;
thisis8
The standardBEM is reasonablyaccurate,but because
of the approximations
when representing
the diffusergeom- H?)(x)• q/2/•xei(x-vz-/2-
z-/4). (9)
etry,notasaccurateasthethinpanelBEI•/[. 4 Thestandard Substitutingthe appropriateGreen'sfunctioninto Eq. (7),
BEM is, however, much faster for two reasons. First, the
neglecting
termsin 1/r2, andu'•ingfar-fieldassumptions
number of elementsrequired to representthe diffuser is
yields
much smaller.Second,becausethe approximatebox geom-
etry of the diffuserstaysthe same during the optimization -A [cos(0) + cos(a)]
process,it is only necessaryto calculatethe surfaceintegrals es(r) =
which determine the element-element interactions once.
8x/lrllr01

C. Predicting the scattering from stepped diffusers XfSR(rs)e-ikxs[sin(a)+s

dxs' (10)
Two predictionmethodswere also available for the whereA givesa measureof the sourcestrength,x s is the x
steppeddiffuser.In this case,however,bothutilized the stan- componentof the vectorr s, and a and 0 are the sourceand
dard BEM. receiver,anglesto the surface'snormal.
(i) The first method exactly models the diffuser's Equation(10) and its derivationshow the limitations
surface--thisis possiblebecauseno fins arepresent.Predict- behindSchroeder'soriginaldesignand the reasonswhy op-
ing the scatteringfrom an arbitraryshapedrigid surfaceby timum diffusionis not produced.This equationdoesrepre-
BEMs is a well established, accuratetechnique. sent a Fourier transformof the surface,exceptfor terms in
(ii) The secondmethodusedan approximate represen- sin(0)+sin(a). Only withoutthesetermswill the Fraunhofer
tation of the diffuser,utilizing a simple phasechangelocal theoryyield constantscattering.whateverthe receiverangle.
reactingadmittanceas was usedfor the diffuserswith fins. The:aretwo furthermajorapproximations: (i) The quadratic
Wherethesurfacegeometry wassuchthattherewasa well residuesequenceshouldbe repeatedto infinity.The trunca-
in thesurface,
thlswasreplaced
by a planesurface
witha tion necessary
for realisticfinite diffuserswill affectthe pro-
phasechangeadmittance at thetop.This r•ducedthenumber ductionof optimumdiffusionbecauseof edgeeffects.(ii)
of surfaceelementsby up to -10% and so reducedcalcula- Equation(10) is only an approximate formulation. For these
tiontime.Testsshowed
thisto induce
a smallbutacceptable reasonsSchroederdiffusersfail to produceoptimumdiffu-
error. sion.

2930 J. Acoust. Soc. Am., Vol. 97, No. 5, Pt. 1, May 1995 Trevor J. Cox: The optimizationof profileddiffusers 2930

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 146.189.194.69 On: Fri, 19 Dec 2014 23:56:07
 III. OPTIMUM DIFFUSER DESIGN
10
The new processto produce optimum diffusers was
basedon an iterativeprocess:
(1) A diffuserwas constructedwith a randomlydetermined
depth sequence.
(2) The pressurescatteredfrom the diffuserwas calculated
usingone of the BEMs.
(3) The scattered pressure
distribution
wasusedto calculate
a parameter whichmeasured thedegreeof diffusion(the
parameter is described
below). -20
(4) The well depthswerealteredaccording to standard
mini- -90 -60 -30 0 30 60 90
mizationtechniques.
(5) Steps(2)-(4) wererepeateduntil a minimumin the dif- Receiver angle (degrees)
fusion parameterwas found indicatingoptimumdiffu-
sion.
FIG. 3. Scatteringfrom a real surfaceat variousfrequencieswith standard
Two standardminimization techniqueswere used: a errordiffusionparameterfor eachfrequencymarked.
simplebutrobust
downhill
simplex
method
•2 anda more
sophisticated
quasi-Newton
method.
•3 Bothmethodsonly
evaluatedat a variety of frequenciesand the resultssummed
use the diffusionparameter'svalue, its derivativenot being
with appropriateweightsdependingon the frequencyrange
known. It was necessaryto run the optimization process where maximum diffusion is desired.
many timeswith differentstartingconditions.The reasonfor Two refinements of the standard error were found nec-
this is that the minimization was being carried out within
essary:
boundedspace.The spaceheld many local minima within
(i) It was necessaryto smooththe scatteredpressure
which the minimizationroutinescouldbecometrapped.The
distributionbefore calculatingthe standarderror. The dif-
solutionspresentedhere are the best found from many at-
fracted distribution had a large number of minima and
temptsof the iterationprocess.This does not exclude the
maximawhich naturallyproducea large standarderror.Over
possibilitythat from some particularstartingpoint yet un-
a wider frequencyrange,say 1/3 octavebands,thesemaxima
tried theremight be a betterminimum achievable.
and minima would smoothout and becomeless significant.
A. Diffusion parameter Furthermore,theoreticalpredictionscan producevery sharp
well pronouncedminima which also increasethe standard
For suchan optimizationprocessto work a suitabledif-
error--such minima would be much less significantin prac-
fusionparameterhad to be developed.This parameterhasto
tice. Smoothingof the angulardistributionwas done;the I o
reduce the large number of resultsfor pressureversus re-
in Eq. (11) werean averageoverten receiversevenlyspread
ceiver positionto a single measureof diffusion.A measure
based on a 95% confidence limit standard error ß was over ten deg. (Smoothingover a frequencyrangewas not
usedbecausepredictingat a variety of frequencieswith the
deemedappropriate:
boundaryelementmethods usedhereis computerintensive.)
(ii) A simpleaverageof the standarderrorsfor many
-
ß- In(
10)•
0 0:-90 x/9ø / • (lo-]o)2 [n(n-1)], (11)
frequenciesallowedany poor frequencyrangesfor diffusion
to be compensated for by othergoodfrequencyranges.This
is obviouslyunsatisfactoryas uniform diffusionis required
where
I ois theintensity
at someangleof reflection
0,•0 is over the entire bandwidth.To prevent such compensation
the averagevalue over 180ø, andn is the numberof samples from occurring,after the standarderrorswere averagedover
in 180ø. Using this standarderror formulationto measure the frequenciestested, one standarderror of the standard
diffusion
is similarto theconcept
suggested
by Schroeder
• errorswas added.This penalizedsolutionswhere different
which was to use "the standard deviation in decibels of the frequencieshadgreatlydifferentdiffusioncharacteristics.
So
energyfluxes."Equation(11) calculatesthe standarderror thenewdiffusionparameterß' for n frequencies
eachhaving
via the intensitiesand so penalizesnonuniformdiffusion a standard
errorof ß//was
more than if decibel values were used. The formulation con-
vertsthe final value to decibelsand representsa 95% confi-
dence limit calculation.
A small value of the standarderror indicatedgood uni-
ß'=•+x/n(n-
1) (ß//_
•)2
form diffusion, a large value poor diffusion.The ideal of (12)
i n
completeuniform scatteringinto all angleswould have a
diffusionparameterof 0. If anotherscatteringdistribution n
i=1
other than uniform is required, the intensities can be
weightedby the inverseof the desiredfunctionbefore the The success of the diffusion measure is illustrated in
calculation of the standard error. The standard error can be Fig. 3. The scatteredpressuredistributionsfrom a real sur-

2931 J. Acoust.Soc. Am., Vol. 97, No. 5, Pt. 1, May 1995 Trevor J. Cox: The optimizationof profileddiffusers 2931

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 146.189.194.69 On: Fri, 19 Dec 2014 23:56:07
 TABLE I. The diffusergeometries.

New Well Well

optimized width Depths Schroeder width Depths(mm)

diffusers (mm) (mm) diffusers (mm) one periodonly

N=7 61 34 154 0 101 0 154 N=7 61 0 50 200 100 100

diffuser O34 200 50
with fins

N=7 61 168 20 55 48 55 20 N=17 63 0 13 50 113 200

stepped 168 100 25 188 163

diffuser 163 188 25 100
200 113 50 13

N=36 59 88 188 200 111 96 N-37 58 0 6 22 50

diffuser 151 127 33 56 133 89 139 200 67
with fins 108 38 23 80 152 150 39 144 56 183
126 170 5 170 126 117 61 17 189 167
152 80 23 38 108 156 156 167 189
133 56 33 127 151 17 61 117 183 56
96 111 200 188 88 144 39 150 67
200 139 89 50 22
6

N=36 59 44 100 1 199 199

stepped 184 69 128 113 97

diffuser 173 142 87 13 1
42 50 39 50 42
1 13 87 142 173
97 113 128 69 184
199 199 1 100 44

face at variousfrequenciesare shownalongwith the mea- times. This meant for diffusers over a certain width, or where
suredstandard errorcalculatedfrom Eqs.(tt) and(12). The therewas a largenumberof degreesof freedomin the mini-
standarderror monitorsthe diffusionsatisfactorily.A disad- mizationprocess,the predictiontimesfor the mostaccurate
vantagein applyingthis diffusionparameteris that it is not methodsbecomeprohibitivelylong. Consequently, whereas
known how big a differencein diffusioncan be heard by the N= 7 diffuserscouldbe optimizedusingthe mostaccu-
listeners. ratepredictiontechniques, theN = 36 diffusershadto utilize
the more approximatemethods.For the final evaluationof
B. Diffusers tested the scattering,and for all plotsgiven in this paper,the most
accurate BEMs have been used.
The diffuserdesignswere constrainedto configurations
This paperhas been restrictedto a singlesourcefrom
which are usedin practiseandwhich did not haveexcessive
oneangleof incidence.One of the primaryusesof diffusers
absorption.Consequently, the limitationson well width and
in auditoriais preventechoes,image shift, and coloration
well depthsuggested by D'Antonio and Konnerthavebeen
adapted; TMtheyhavea largeamount ofexperience ofproduc- from strongfirst-orderreflections.In that caseit seemsrea-
sonableto deal with a single angle of incidencefor sound
ing Schroederdiffusers.A well width of about6 cm and a
maximumwell depthof 20 cm havebeenused. coming straightfrom the stage.(Ideally a smallrangeof
Four differentconfigurations
have beentested: incidenceanglesshouldbe includedto allow for the varia-
tion in incidenceanglesdueto the stagewidth anddepth.)
(a) diffuserwith fins, sevenwells, Althoughthe diffusionwas only optimizedfor one angleof
(b) diffuserwith fins, approximately36 wells, incidence,the diffuserswill still providea certainamountof
(c) steppeddiffuser,sevenwells, and diffusionfor higher-orderreflectionsfrom other anglesof
(d) steppeddiffuser,approximately 36 wells. incidence.There is no reasonwhy a rangeof incidentangles
All these configurationshave been compared to couldnot be usedto try and obtainoptimumdiffusioninde-
Schroederdiffuserswith similar geometries.Full details of pendent of incidentangle.This would, however,slow the
the diffuserdesignscanbe foundin TableI. This includesthe optimization process.
well depthsof the best diffusers.The sourcewas a normal A normal sourcewas usedin this projectas it allowed
distance of t0 m from the diffuser, the receiver at a fixed the exploitationof symmetry,greatly reducingprediction
radiusof 5 m, and the overall depthof all diffuserswas 0.3 timesand halvingthe degreesof freedomin the minimiza-
m. tion process.(The predictions for the Schroederdiffusers,
Ideally,the mostaccurateBEMs would havebeenused however, were carriedout with asymmetrical surfaceswhere
duringall the optimizationprocesses. Unfortunately,during appropriate.)
optimizationthe scatteringhasto be evaluatedthousands of The work hasbeenrestrictedto a singlereceiverradius.

2932 J. Acoust.Soc.Am., Vol. 97, No. 5, Pt. 1, May 1995 TrevorJ. Cox:The optimization
of profileddiffusers 2932

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 146.189.194.69 On: Fri, 19 Dec 2014 23:56:07
 -5

-10

-15
0 250 500 750 1000 1250 -90 -60 -30 0 30 60 90

Frequency(Hz) Receiver ange (degrees)

FIG. 4. Comparisonof diffusionfrom: -- optimizeddiffuserwith fins, FIG. 5. Comparisonof scatteringfrom N= 7 Schroeder diffuser,
N-7 Schroederdiffuser,and ...... optimizedsteppeddiffuser.Dif- ..... optimizeddiffuserwith fins, and• optimizedsteppeddiffuserat
fusioncharacterizedby standarderror parameter. 250 Hz.

When this methodis appliedto a real problem,the rangeof wells with the sameoverall width. The following were tested
receiveranglesand distancesencounteredin the actualaudi- (periods
X prime number):5X7, 2X17, and 1X37. These
torium oughtto be used. have design frequenciesof 490, 808, and 834 Hz, respec-
The low-frequency performanceof the diffusers has tively. In Figs. 7 and 8, comparisonsbetweenthe various
beentested(up to 1250Hz). At thesefrequencies,thediffus- Schroeder diffusers and the best optimized diffuser are
ers couldbe usedto improvethe diffusionin a hall, increas- shown.The optimizeddiffuserwas betterthanthe Schroeder
ing the amountof early lateral energyand so increasingthe diffusers,althoughthere is only a small increasein perfor-
sensationof spatial impression.Producingdiffuserswhich mancewhen comparedto the 2X17 diffuser.
operateto higherfrequencies mightbestbe achievedby con- As the designtheory behind Schroederdiffusersis only
structing
two-waydiffusers.
TMFor theoptimization,
seven approximate,it was expectedthat the new optimizeddiffus-
randomly chosen frequencieswithin the bandwidth were ers would create significantlybetter diffusion.Yet the im-
used(101, 284, 487, 651, 807, 973, and 1170 Hz). This was provement on the scattering compared to the 2X17
found sufficientto get good diffusion over the bandwidth Schroederdiffuseris small.The failure to produceconsider-
whetherat or away from theseoptimizationfrequencies.The ably better diffusionis mainly due to the fact that the scat-
graphsdisplayedhereare not at the optimizationfrequencies teringfrom Schroederdiffusersis alreadyfairly uniform.It is
and so representthe worstcasesof the new optimumdiffus- difficult to producemore uniform scatteringwithin the con-
ers.
straintsof geometryusedhere.This is not, however,particu-
larly due to the use of the quadraticresidue sequenceto
IV. RESULTS
determinethe well depths.Even diffuserswith randomly
A. Diffuser with fins, seven wells

Diffusersproducedby the optimizationprocessoutper-

formed the N= 7 Schroederdiffuser over a wide variety of
frequencies.This is illustratedin Fig. 4. where the scattering
is characterizedby the standarderror. In Figs. 5 and 6, the
scattering
at two particularfrequencies
are shown.(The de-
signfrequency
for theSchroeder
diffuser
is 490Hz.3) The
optimizeddiffuser'ssuccess is particularlydue to the narrow -10
natureof the SchroederdiffuserviolatingSchroeder'sdesign
theory. One period of N= 7 is not an infinite repeat of a
-20
quadraticresiduesequenceand so the effectsof the edgeand
the back of the diffuserare very noticeable.Consequently,
i
for narrowdiffusersdramaticimprovementsin the scattering -3O i i i • i

can be achievedusingthis optimizationmethod. -90 -60 -30 0 30 60 90

B. Diffusers with fins, approximately 36 wells Receiver ang e (degrees)

Optimizationof a diffuserwith 36 wells was carriedout FIG. 6. Comparisonof scatteringfrom N= 7 Schroeder diffuser,
and compared to a variety of Schroeder diffusers. The ..... optimizeddiffuserwith fins, and • optimizedsteppeddiffuserat
Schroederdiffuserswere designedto have approximately36 1050 Hz.

2933 d. Acoust. Soc. Am., Vol. 97, No. 5, Pt. 1, May 1995 Trevor d. Cox: The optimizationof profileddiffusers 2933

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 146.189.194.69 On: Fri, 19 Dec 2014 23:56:07
 _/

I I I I I

0 250 500 750 1000 1250

0 250 500 750 1000 1250

Frequency(Hz)
Frequency(Hz)

FIG. 7. Comparisonof diffusion from various Schroederdiffusersand an FIG. 9. Comparisonof diffusionfrom .... 2x 17 Schroeder diffuser, •
optimizeddiffuserwith fins. All diffuserswith about 36 wells. 1x37 an optimized diffuser with fins, and a diffuser with random well
Schroeder diffuser, 2x17 Schroeder diffuser; ....... 5x7 depths.
Schroederdiffuser,and • optimizeddiffuserwith fins.

E. Discussion of stepped diffusers

chosendepthsproducereasonablediffusion.This is illus-
tratedin Figs. 9 and 10 where the scatteringfrom the 2x 17 Removing some of the constraintson geometryfrom
andoptimizeddiffuseris comparedto a diffuserwith random Schroeder-type diffusers--byremovingthe fins--allows the
well depths.(This diffuserhastwo periodsof 18 randomly productionof betterdiffusion.This successshowsa possible
chosendepths.) directionfor further study.The methodsoutlinedin this pa-
per enablediffusersto be designedwithoutrigid compliance
C. Stepped diffusers, seven wells to certainconfigurations. This could enablethe acousticde-
siresof acousticiansand visual requirementsof architectsto
In Figs. 4-6 resultsfrom the optimizationare shown. be betterblendedinto an auditoria.The successof stepped
The steppeddiffuseroutperformsboththe Schroederdiffuser diffuserswas also achievedunder arguablyunfair criteria.
and the optimumprofiled diffuserat nearly all frequencies. Diffusers of similar absorptionshould be compared.This
meansthat the steppeddiffuserscould have greaterwell
D. Stepped diffuser, approximately 36 wells depthsfor similar absorptionas the Schroederdiffusers,and
Comparisonsof the scatteringfrom the best Schroeder becauseof this, the possibilityof even betterdiffusion.
diffuser, the optimized diffuser with fins, and the stepped There are two difficulties,however,with steppeddiffus-
diffuserare shownin Figs. 11 and 12. As with the diffusers ers. First of all, there are no simple designequationswhich
with sevenwells, the steppeddiffuserproducesthe bestdif- can easilybe implementedon a desktopPC as there are for
fusion. Schroederdiffusers. Second,for wide diffusersor thosewith
a large number of wells, the optimizationprocesscan be-
10
!0

-10
-10

-20

-20
-90 -60 -30 0 30 60 90
-30
-90 -60 -30 0 30 60 90
Receiver ange (degrees)
Receiver ang e (degrees)
FIG. 8. Comparisonof scatteringfrom variousSchroederdiffusersand an
optimizeddiffuserwith fins at 350 Hz. All diffuserswith about36 wells. FIG. 10. Comparisonof scatteringfrom 2x17 Schroederdiffuser,
..... lx37 Schroeder diffuser, 2x17 Schroeder diffuser, • an optimizeddiffuserwith fins,and..... a diffuserwith randomwell
5 x 7 Schroederdiffuser,and • optimizeddiffuserwith fins. depths(1050 Hz).

2934 J. Acoust. Soc. Am., Vol. 97, No. 5, Pt. 1, May 1995 Trevor J. Cox: The optimizationof profileddiffusers 2934

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 146.189.194.69 On: Fri, 19 Dec 2014 23:56:07
 teredtogether.)
Outsidethedomainof optimization,
theop-
timizedandsteppeddiffuserswerefoundto give roughlythe
samediffusionas the Schroederdiffusers--sometimes the
new diffusershad worse diffusion, sometimesbetter than the
Schroeder diffusers. These results demonstrated that for best
resultsthe diffusersshouldonly be usedundertheconditions
they have been optimized.They also usefullydemonstrate
that good diffusionwithin the optimizedregion has been
achievedwithout overly penalizingthe diffusers'perfor-
mances elsewhere.

v. CONCLUSIONS
250 500 750 1000 1250
Using an iterativeprocessthe performanceof profiled
Frequency(Hz) diffuserscan be optimized.In this studythe productionof
uniformscatteringfor all anglesof reflectionhasbeenstud-
ied.The techniques havethepotentialfor applicationto other
FIG. 11. Comparisonof diffusion from 2x 17 Schroeder diffuser,
an optimizeddiffuserwith fins, and an optimizedsteppeddif-
diffusioncriteriaand other diffusergeometries.To monitor
fuser. the degreeof diffusion a measurebased on standarderror
formulations was used and found to be successful.

comeprohibitivelylong. The productionof more sophisti- Two typesof optimizeddiffusershavebeencomparedto

cated designmethodsor more powerful computerscould Schroeder diffusersbasedon the quadraticresiduesequence.
solve this problemin the future. The first type of optimized diffuser was similar to the
Schroederdiffusersconsistingof a series of wells of the
F. Measurements samewidth,but of differentdepthsseparated by thin fins.In
this casebetterdiffusionwas achievedwith the optimized
A reviewerof this papercarriedout substantialwork to diffuser.The improvementwas greatestfor the narrow dif-
confirmthe resultsof the predictedresponses;their work fuser tested.
included measurements of the three diffusers with seven
The secondtypeof optimizeddiffuserwasessentially a
wells. Thesemeasurements confirmedthe resultspresented Schroederdiffuserwith the fins removed.These"stepped"
above,givingextremelygoodagreementwith the theoretical diffusersoutperformedboth Schroederdiffusersand the first
predictionsandconfirmingthe improveddiffusionproduced type of optimized diffuser with fins. With the lack of fins,
by the new optimizedsurfaces. suchdiffusersshouldbe cheaperto produceandbe lessab-
With these measurements,the reviewer also looked at sorbent than Schroeder diffusers.
the performance of the diffusersoutsidethe domainof opti-
mization;at higherfrequencies,at otheranglesof incidence, ACKNOWLEDGMENT
for differentreceiverradii, andwhenmany of the diffusers
wereclusteredtogether.(The diffuserswere testedfor a fre- The authorwould like to thankthe staff at the Depart-
quencyrangefrom 300 Hz to 3 kHz, 30ø and 60ø anglesof ment of Applied Acoustics,Universityof Salfordfor their
incidence,andwhen two periodsof the diffuserswere clus- help with this work.

10
M. R.Schroeder,
"Binaural
dissimilarity
andoptimum
ceilings
forconcert
halls: More lateral sounddiffusion," J. Acoust. Soc. Am. 65, 958-963
(1979).
2M. R. Schroeder,
"Progress
in architectural
acoustics
andartificial
rever-
beration:Concerthall acoustics
andnumbertheory,"J. Audio. Eng. Soc.
32, 194-203 (1984).
3p.D'Antonio
andJ. Konnert,
"Thereflection
phase
grating:
Design,
theory,andapplication,"J. Audio. Eng. Soc.32, 228-238 (1984).
4T.J. CoxandY. W. Lam,"Prediction
andevaluation
of thescattering
from quadratic residue diffusers," J. Acoust. Soc. Am. 95, 297-305
-2o \ (•994).
5B.A. deJongandP.M. vandenBerg,"Theoretical
design
of optimum
planarsounddiffusers,"J. Acoust.Soc.Am. 68, 1!54-1159 (1980).
-6o -so o so 6o aT.Cox,"Predicting
thescattering
fromreflectors
anddiffusers
using
two-
dimensionalboundaryelementmethods,"J. Acoust. Soc. Am. 96, 874-
878 (1994).
Receiver angle (dB) 7T.Terai,"Onthecalculation
offields
around
three-dimensional
objects
by
}ntegral
equation
methods,"
J.Sound.
Vib.69,71-100(1980).
FIG.12. Coml•arison
of scattering
from 2x17 Schroeder
diffuser, M. Abramowitz
andI. A. Stegun,
Handbook
ofMathematical
Functions
-- an optimizeddiffuserwith fins, and..... an optimizedsteppeddif- (Dover, New York, 1965).
fuser (250 Hz). 9y. KawaiandT. Terai,"Theapplication
of integral
equation
methods
to

2935 J. Acoust.Soc.Am.,Vol.97, No. 5, Pt. 1, May 1995 TrevorJ. Cox:The optimization

of profileddiffusers 2935

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 146.189.194.69 On: Fri, 19 Dec 2014 23:56:07
 the calculation
of soundattenuation
by barriers,"Appl.Acoust.31, !01- 12W.H. Presset al., NumericalRecipes,theArt of Scientific
Computing
117 (1990). (CambridgeU.P.,Cambridge, 1989),pp.289-292.
•øA. J. Burton,"The solutionof Helmholtzequationin exteriordomains NAGroutines librarymanual,Numerical Algorithms GroupLtd.,Oxford,
UK.
usingintegralequations,"
NationalPhysicalLaboratoryReportNAC 30,
1973. •4p.D'AntonioandJ. Konnert,"The QRD diffractal:A new one- or two-
•H. A. Schenck,"Improvedintegralformulation
for acoustic
radiation dimensionalfractal sounddiffusor,"J. Audio. Eng. Soc. 40, 117-129
problems,"
J.Acoust.
Soc.Am.44,71-100(1968). (1992).

2936 J. Acoust.Soc.Am.,Vol.97, No.5, Pt. 1, May1995 TrevorJ. Cox:The optimization

of profiled
diffusers2936

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 146.189.194.69 On: Fri, 19 Dec 2014 23:56:07