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

net/publication/200045074

Image method for efficiently simulating small-room acoustics

Article in The Journal of the Acoustical Society of America · April 1979

DOI: 10.1121/1.382599

CITATIONS READS

3,159 2,973

2 authors, including:

Jont B. Allen
University of Illinois, Urbana-Champaign
242 PUBLICATIONS 10,122 CITATIONS

SEE PROFILE

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

Speech feature identification View project

All content following this page was uploaded by Jont B. Allen on 07 December 2014.

The user has requested enhancement of the downloaded file.

Image method for efficiently simulating small-room acoustics
Jont B. Allen and David A. Berkley
.4cousties
Research
Department,
Bell Laboratories.
MurrayHill, NewJersey07974
(Received6 June 1978)

Imagemethods are commonlyusedfor the analysisof the acousticpropertiesof enclosures.In thispaper

we discuss the theoreticaland practicaluseof imagetechniquesfor simulating.
on a digitalcomputer,the
impulseresponse betweentwo pointsin a smallrectangular room.The resultingimpulseresponse, when
convolved withanydesired inputsignal,suchasspeech, simulates
roomreverberation of theinputsignal.
This!echnique is usefulin signalprocessing
or psychoacoustic
studies.The entireprocessis carriedouton
a digitalcomputer sothata widerangeof roomparameters canbestudiedwithaccurate controloverthe
experimental conditions. A FORmS• implementationof thismodelhasbeenincluded.

PACS numbers: 43.55.Ka, 43.55.Br

INTRODUCTION
(1) We are most interested in the office environment,
which is usually a rectangular geometry.
In some recent experiments, which studied the per-
(2) This model can be most easily realized in an ef-
ceptual effects of reverberation properties of a small
ficient computer program.
room,t.2 a carefullycontrolled,easilychanged,acou-
stic environment was required. It was decided to utilize (3) The image solution of a rectangular enclosure
a computer simulation of the acoustic space. This pa- rapid/y approaches an exact solution of the wave equa-
per describes both the general theoretical approach and tion as the walls of the room become rigid.
the specific implementation techniques used (the
The image model is chosen because we are interested
FORTRANprogram). We believe that the resulting room
in the point-to-point (e.g., talker-to-microphone) trans-
model is useful for a broad range of investigations,
fer function of the room. In order to obtain a good
from our original experiments mentioned above, to
trausientdescriptionofthe response, atimedomaln mo-
basic studies of room acoustics.
del is required. A normal-mode solution of the enclosure
The room model assumed is a rectangular enclosure would require calculation of all modes within the fre-
with a source-to-receiver impulse response, or trans- quencyrange of interest (i.e., 0.1-4.0 kHz), plus cor-
fer function, calculated using a time-domain image ex- rections for those outside this range. The image meth-
pansionmethod. Frequent applications have been made od includes only those images contributing to the im-
of the image method in the past as in deriving the re- pulseresponse. Thus the contributingimagesare those
verberation-time
equations,
3 for theoreticalstudies within a radius given by the speed of soundtimes the
soundbehaviorin enclosures,4-?andin the studyof reverberationtime.• (The exact relationshipbetween
architectural acoustics and perceptual properties of the normal-mode solutions and the image solution, for
rooms?'n In addition,there hasbeena considerable a lossless room, is discussedin AppendixA.) The im-
amount of important theoretical work on the approxi- portant information used here is that in the time-do-
mate12useof imagesproduced
by a singlesoft-wall main, each image contributesonly a pure impulse of
(finite impedance) reflection. Several recent papers on knownstrengthand delay while each normal mode is
this subject which have goodbibliographies are Refs. a decayingexponentialwhich contributes to all times.
13, 14, and 15. Computer methods have also recently Furthermore, whereas an image has only delay and
beenapplied to image computationsin enclosures(see gain as parameters, a normal mode computationre-
for example Refs. 6, 7, 10, and 11). In the current quires the solutionof transcendentalequationsto find
paper the computational technique is specifically aimed the Dotelocation plus the evaluation of a relatively
at being simple, easy to use, and fast. In addition the comptexfunctionto find the modegain (i.e., the residue
resulting room responses have been used to realistical- of the pole).
ly model speechtransmission in rooms and to investi-
gate the effects of various forms of digital speech sig-
ns/ processors.
16'tl A. The image model
We model a talker in a room as a point source in a
In the following we will first briefly discuss theoretic-
al aspectsof the method. Thenwe will outlinethe compu- rectangular cavity. A single frequencypoint source of
tational approach and, finally, we will give some acceleration in free space emits'a pressure wave of the
form
staples of applications.

I. IMAGE VERSUS NORMAL MODE MODELS P......

We model the rooms of interest as simple rectangular
enclosures. This choice of geometry is made for sev-
eral reasons: where

943 .I.Acoust.
Soc.Am.65(4),Apr.1979 0001-4966/79/040943-08500.80¸1979Acoustical
Society
ofAmerica 943
 IMAGE EXPANSION p(t}
P = pressure,
Y

f= frequency,
o

t= time,

R-- Ix-x'l, (2)

X =vector talker location (x,y,z),
X' =vector microphonelocation (x', y ', z'),
p(1)=•.. •..,' .•
i= v•-i, p=lr=-<•4w'IRp+
Rrl
c = speed of sound.
•r=(2nLx,21Ly,2mLz)
When a rigid wall is present, the rigid wail (zero norm-
al velocity) boundary condition may be satisfied by •=(x• x',y•y',z•z')
placing an image symmetrically on the far side of the
wall. Thus, •p+;r =(x•x'+2nLx,•
t y'+21Ly,z•z'+2mLz)
YTG. •. A slice •u•h the t•;e space show• how •e
p(w
,X,X')=[oxp[i4(•R/.c)R*]
+exp[i4(•/.c)R"]]exp(-icot),
(s)
i•ges of the source are spatially arra•ed.
represen• •e origi•l more. The ac•
The solid •x
image s•ce is
•me dimensional.

where we define the two distances from the microphone

to the source R. and to the image R. by
=(x_ x,)2 _y,)2+(z _ (4) be derived directly from the normal-mode solution as
2.=(x + gF +(y +(z -z') 2ß shown in Appendix A.
The wall has been placed at x = 0 in this case (note the
B. Caseof nonriõid wslls
sign in the x terms of R. and R.).
In the general case of six walls the situation becomes If the room walls are not rigid, the solution in terms
more complicated because each image is itself imaged. of point imagesmay no longe• be exact. A precise•
The pressure may then be written (as shown in Appen- statement of the effects of finite impedance walls is
dix A) presently impossible, since the effects on even a single
imageare quitecomplicated.
ts'•4'• Thereforewehave
continued to assume the approximate point image model
even for nonrigid walls. In addition we have assumed
an angle independent pressure wall reflection coefficient
where R• represents the eight vectors given by the eight •. This assumption is equivalent to assuming that the
permutations over ñ of wall impedance is proportional to sec(8), where 8 is the
l•=(x•x', y •y', z•z') (6) angle of incidence of a plane wave with respect to the
wall normal. We presently do not understand the exact
r is the integer vector triplet (n,l,m), and physical interpretation of the above assumptions. How-
lL, mr,), (7) ever, we believe that they do not introduce serious
problems into the final result under typical conditions.
where (L•, L•, L•) are the room dimensions. Equation By typical, we mean over the frequency range of 100
(5) is the pressure frequency response assuming rigid Hz-4• kHz, wall reflection coefficients of greater than
walls for a point source at X =(x,y,z) and receiver at 0.7, typical office room geometries, and where both
X'=(• ,y',z'). If Eq. (5) is Fourier transformed, we source and receiver are not close to the wall. Many,
find the room impulse response function (time domain if not all, of the above conditions are probably not cri-
Green's function) tical and could be relaxed. We merely wish to carefully
pointout the nonexactnatureof the •esults.

p(t,X,X'
=•_.•,•.• 4vli•+RrI The above assumptions result in the Sabine energy
absorption coefficient c• for a uniform reflection coeffi-
cient fi on a given wall of the form
An interpretation of Eq. (8) is given in Fig. 1 where we
show a part of the image space for a two-dimensional a = 1 - •2. (9)
slice through the room. When the accelerative source
Our assumptionsare similar to thoseof geometrical
location (talker)X is excited, each image point is si-
multaneously excited, creating spherical pressure
acoustics
• andare the sameas thoserequiredfor spec-
ular angle-independent ray tracing. In current imple-
waves which propagate away from each image point.
mentations of the model we also do •ot allow frequency
Equation (8) is the exact solution to the wave equation variations in the reflection coefficients. Both the angie
in a rectangular, rigid-wall (lossless), room and may dependenceand frequency dependence could be included

944 J. Acoust.Soc.Am.,Vol. 65, No.4, April 1979 J. AllenandD. Rerkley:Methodfor simulating

small-room
acoustics 944
inourcomputer
program,
b•t onlyat theexpense
of TABLE I. Computationparameters•room size (feet) 10' x 15'
significantly complicating and slowing down the compu- x 12.• 8 kHz samplingrate.
tational model.
Impulse response Convolution
Introducing the effects of finite, angle independent Leugth Image Computation rate
(ms) No. points count time (s) (s/s)
wall absorption into Eq. (8) leads to the modified room
impulse response 64 512 585 1 12.5
128 1024 469 0 8 13.8
256 2048 375 00 60 15.0
p(t,X,X')-z.• z_,•,t ,• • •2•,•

x
4.,,.
fR,+ ' (10)
where 1• is now expressed in terms of the integer 3-
vector p---(q,j,/•) as I for our implementation on a Data General, Eclipse
P,•=(x-x' +Zqx', y -y' + 2jy', z-z' + 2kz'). (11) S/200 computer. (On this machine the computation
time required for each image is about1.6 ms.) T.he
I• as givenby Eq. (6) is similar to titat of (11), but is actual FORTRANprograms used are given in Appendix B.
indexeddifferently from (11). The beta's are the pres-
sure reflection coefficients of the six boundary planes, The temporal quantization in the impulse function
with the subscript I referring to walls adjacent to the computationcauses sligh[ statistical errors in the com-
coordinate origin (see Fig. 1). Subscript 9. is the oppos- putedarrival times of each image pulse relative [o the
ing wall. Eq. (10) has been derived heuristically from exact delay as given in Eq. (10). This error can be
geometrical
considerations
of Fig. 1. Thesum• with thoughtof as effectively "moving" each image source by
vector index p is used to indicate three sums, namely 0•< e•-•'o•'•< AR/2 relative to the receiver. This effect
one for each of the three componentsof p=(q,j,k). could be removed, in principle, by using a band-Limited
r--(n,l,rn) is a similar sum. Physically these sums source pulse. However, the error is small for most
are over a three-dimensional latUce of points. For p (if not all) purposes and it greatly complicates the com-
there are eight points in the lattice and for r, the lattice putation to remove this approximation. We have esti-
is infinite. mated that the error due to the slight moving of the im-
ages could not be perceived even in a digital simulation
of a binaural hearing experiment.
II. IMPLEMENTATION OF THE MODEL
The subroutine SROOM of Appendix B requires as pa-
The primary considerationin a computer (sampled rmeters the number of impulse response points de-
data) implementation of Eq. (10) is the method of spatial sired (NPTS), the source location R0, the receiver 1o-
sampling. In addition, an apparently nonphysical be-
havior of the model at zero frequency is removed by a
low-frequency (0.01 of the sampling frequency) high-
passdigitalfilter.19
1.0
A calculatedimpulse responseis built up as a "histo-
gram" oi image pulses received at different time de-
lays. The width of each histogram bin is equal to the IMPULSE RESPONSE

time sampling period T initially assumed, which in

turn is determined by the highest frequency to be rep-
resented. For example, all images with the range
N•R to (N+ 1)aR, where AR=½T (T is the sampling
period and c the speed of sound), are added together
with appropriate amplitude as given by Eq. (10).
The choice of sampling rate is governed.by the appli-
cation. If speech is to be studied in small rooms one
might choose T = 0.1 ms. (sampLing frequency of 10
kHz; highest frequency of 5 kHz). But, if reverberation
times of large enclosures are being studied (and convo-
2048 POINTS
lution with speechis not required) much lower rates 8 KHZ SAMI:4.1NG RATE
can be useful.
-I.0 I I I I I I I I
0
The time length of the calculated impulse response is TIME (MS)
also a consideration. For a given sampling rate the
numberof pointsin p(•) increases linearly with its FIG. 2. Plot of a typical impulse response for a room
length while the computationtime (and number of im- 80 x 120 x 100 sample lengths long. Wall reflection coefficients
ages) goes up approximately as the cube of response were all 0.9 ceiling and floor coefficients were 0.7. X and X'
length. This is shown in the first four columns of Table were at (30, 100, 40) and (50, 10, 50) sample periods.

945 J. Acoust.
Soc.Am.,VoL65. No.4, April1979 J. AllenandD. Berkley:Methodfor simulating
•mall-roorn
acoustics 945
cation R, the room dimensions, all specified in terms usea modificationof the integratedtone-burstmethod
22
of the sample length (AR), and the reflection coefficients
of each of the six wall surfaces (fi). Figure 2 shows an
example of the impulse response obtained for a room of
z(t)=kf (12)
dimensions 80 x 120 x 100 sample lengths with equal wall where E(t) is the average energy decay, k is a propor-
reflection coefficients of 0.9 ((• = 0.19) and with floor tionality constant, and p(?) is the calculated pressure
andceilingreflectioncoefficients(/],) of 0.7 ((• = 0.51). X impulse response from Eq. (10). For cases where the
and X / were (30,100,40) and (50, 10,60) sample in- impulse response has been truncated before most of the
tervals, respectively. decay has taken place, (12) may lead to errors. These
errors are usually obvious in the E(t) plots.
It is usually convenient to interpret the model param-
eters as a true distance rather than as multiples of Another, approximate, approach is to simply mea-
AR. This requires the choice of a sampling rate and sure the short-time average energy decay of the im-
then conversions may be performed in the users main pulse itself (e.g., using a simulated level recorder).
program which calls the subroutines of Appendix B. For exponential or near-exponential decays, both
Figure 2 is labeled assuming an 8 kHz sampling rate. methods should give approximately the same value of
For this assumption (and assuming a sound speed of 1
ft/ms) the room dimensions are 10'x 15' x12.5'.

III. APPLICATIONS

Our room image model has been applied to several (o)

problems. We will discuss t•vo examples: a psycho-
physical
evaluation
of roomreverberation
effects
t'2and
a study of critical distance measurements using spec-
tral response
variance.
2øWe havealsousedthe model
to test a signal processor intendedto reduce perceived
reverberationt6 andto studyproblemsassociatedwith
mathematical inversion (inverse filtering) of room
transfer functions.t? DECAY CURVE

A. Psychophysicsof room reverberation

Once a simulated room impulse response has been
calculated using the image model, the psychophysical
effects of this simulated reverberation on speech may
2048 POINTS
be directly studied. A reverberant sample of speech 8 KHZ SAMPLING RATE
was producedby convolving an anechoic (unreverberant)
speech sample with the calculated impulse response I
256
[p(t)]. This can be doneefficiently usinga Fast Fourier TIME (MS)
Transform (FFT) method (overlap-add) to perform the
convolution.2• The last column of Table I shows the
measured convolution rate, for various length impulse
responses. The convolution rate only increases as
log2(N), (where N is the room response length in time
sample periods T) so even large impulse responses can
be convolved with speech quite efficiently. For exam-
ple, to convolve (filter) one secondof speech, sampled
at 8 kHz, with a 256 ms long impulse response (2048
points) requires a 15 s computation.
The speed of processing makes multivariate psycho-

ENERG
DECA
CURV
physical studies quite practical. Ease of modification
and perfect control of room parameters avoids the
problems which have made such experiments so difficult in
the past. The actual experiments used 16 different simu-
lated" rooms" (impulse responses)convolvedwith ten dif-
ferent sentences spokenby four different speakers. Itwas 2048 POINTS
discovered that the experimental rooms were perceptually 8 KHZ. SAMPLING RATE

well characterized by their spectral variance [Eq. (14)]

0 256
andby the reverberation time. This latter measure, TIME (MS)
reverberation time, deserves some discussion.
FIG. 3. (a) Energy decay curve for the impulse response of
Given the impulse responses, reverberation time may Fig. 2 usingthe Schroeder[ntegraUon
method.•2 (b) Impulse
be estimated in a number of ways. One method is to energy decay curve for a simulated level recorder.

946 J. Acoust.Soc.Am., Vol. 65, No. 4, April 1979 J. Allen and D. Berkley: Methodfor simulatingsmall-roomacoustics 946
reverberation time. Example plots of E(t) for both empirically that calculated reverberation times, for a
methodsare shownin Figs. 3(a) and 3(b) using the im- number of simulated enclosures, agree well with
pulse responseof Fig. 2. ExperienceindicatesthatEq. Eyring'sformula
• overa widerangeof Betavalues?
(12) gives the most satisfactory results. We have found
In the experimentsdiscussedabove1,24wediscovereda
monotonic
relationship
between
A/A:( Fig.4)themicro-
phone-talker distance when normaliz edby the room c ri-
(a) STANDARD
DEVIATION
OFSPECTRAL
RESPONSE tical distance(thedistanceat whichreverberant energy
IN COMPUTER-SIMULATED
ROOM:17' x I$' x I0'
7
equalsdirect soundenergy), the reverberation time, and
psychophysical preference for the resulting speech.

o
B. Critical distance measurement
ß o
&• A A A newmethodhasbeenproposed
2øfor measurement
G o of critical distances (or reverberation radius) in rooms.
In this techniquea measurementis madeof the log frequen-
cy responsevariance cL defined as
L(co):201og[[p()I
] (13)
c[ =[œ(•)-L(,•)]• , (14)
given the room transfer-functionP(co)[Fourier trans-
form of Eq. (10)] for several microphone-sourcespac-
0.15 ings. The measured values are fitted to a theoretical
029
050 curve for aL based on the assumption of simultaneously
•070 excited, uncorrelated, normal modes, combined with
the calculated direct sound energy. The resulting fit
was shown to give an accurate value for the room's
critical distance.

This new method was extensively studied using our

I I I image model before being applied successfully to real
-30 -24 -18 -12 -6 0 6 12 18 24
rooms. Since the direct and reverberant energy are
REVERBERANT/DIRECT ENERSY RATIO
(A/Ac)• known in the computer model, a comparison can easily
be made to the theory. The model results show excell-
(D)STANDARD
DEVIATION
OFSPECTRAL
RESPONSE ent agreement with theoretical calculation as is seen in
IN COMPUTER-SIMULATEDROOMi47' x 31' x
Figs. 4(a) and 4(b). We know of no other method by
which this study could have been carried out as effect-
ively.

IV. SUMMARY ANO DISCUSSION

A simulation method for small rooms based on an

approximate image expansion for rectangular nonrigid-
wall enclosures has been discussed. The method is
simple, easy to implement and efficient for computer
simulation. Several examples of its use, where other
methods would be difficult, have been discussed.

APPENDIX A
• a 0.15
)• a 0.29 We wish to derive the rigid-wail image solution di-
}• * 0.50 rectly from the normal-mode expansion for a rectangu-
y •0. 70 lar enclosure. The frequency response function
(Green's function) for the pressure P(co) in an enclos-

-30
J ROOM
-P4
DIMENSIONS:
47'x13'x
15'
-18 -12 -6 0 6 t2 18 24
ure is given by solving the Helmhoitz equation driven
by a single frequency point acceleration source.

vP[(co/c),X,x'] + ,x'] =- -x '),

REVERBERANT/ DIRECT ENERGY RATIO (dB)

( A/A c)2 (A1)

FIG. 4. Figuresfrom Jetztzøwhichcomparethe theoretical
rms deviation of the pressure in dB from the mean pressure where co is the frequency and c is the speed of sound.
in dB as a function of the direct to reverberent energy ratio The solution to this equation, assuming rigid boundar-
(a) for a room 17x13x10 ft and (b) 47x31x15 ft, ies, is given by

947 J. Acoust. Soc. Am., Vol. 65, No. 4, April 1979 J. Allen and D. Berkle¾: Method for simulatingsmall-roomacoustics 947
 ..,• __ • *r(x)*r(x ) By Fourier series analysis one may show
P(k,X,,,j- /•
V,=.. 2 k2)
(k,- ' (A2)
where k = •/c, r = (n, l,m) indicates a three dimension-
al sum, V is the room volume,
Thus[withanalogus
equations
to(A9)fory andz]
•=• z• .

k2r=]kr12 (A3)
p(•,X,X,)=(21_•)
•_f f f •-•exp[(i•.(R•+
P•)]
(A•0)
and
where• is thevector[alsogivenby Eq. (7)]
n•rx l•y rn•rz
•=2(•, zL•, m•,). (An)
where the Li•s are the room dimensions. Each triple integral is just a plane wave expansion for
a •int source in free space since
Using the exponential expansion for cosine, mMti-
plyi• the terms of Eq. (A2) together and collecti•, we
obtain •p(iklRI)I f•f exp(i•.R)
• (A12)
x • •exp(,•.•) Finally, •ing Eq. (A12), Eq. (A10) becomes
P(k,X,X')=•,• (•$_•:), (AS)
where • representsthe eight vectors [also givenby

•=(x•x •, y •y•, • •'). (A6)

T•ing theinverseFouriertransform
ofEq.(A18),the
Usingtheproperty of thedelta Mnctiononkz, k• • andk• echo structure becomes explicit

f• •(•- =)•(•)•
=F(=), (A•)
we may rewrite Eq. (AS) in integr• form ,• 4•[•+•} ' (A14)

which is the same as Eq. (8) as desired.

APPENDIX B

C PGM: SROOM
C SUBROUTINE TO CALCULATE A ROOM IMPULSE RESPONSE
C R=VECTOR RADIUS TO RECEIVER IN SAMPLE PERIODS =LENGTH/(C*T)
C R0 =VECTOR RADIUS TO SOURCE IN SAMPLE PERIODS
C RL =VECTOR OF BOX DIMENSIONS IN SAMPLE PERIODS
C BETA=VECTOR OF SIX WALL REFLECTION COEFS (0 <BETA <=1)
C HT=IMPULSE RESP ARRAY
C NPTS=# OF POINTS OF HT TO BE COMPUTED
C ZERO DELAY IS IN HT(1)
C
SUBROUTINE SROOM(R, R0, RL, BETA, HT, NI•TS)
DIMENSION HT(NPTS)
DIMENSION R(3), R0(3), NR(3), RL(3), DELP(8), BETA(2,3)
EQUIVALENCE (NR(1), NX), (NR(2), IVY), (NR(3), NZ)
DO 5 I=l, NPTS
5 HT(I) =0
C CK FOR MIC AND SOURCE AT SAME LOCATION
DIS = 0
DO 6 I=1,3.
6 DIS = (R(1)-R0 (I))*'2 +DIS
DIS = SQRT(DIS)
IF (DIS.LT..5) HT (1) = 1
IF(DIS.LT..5)RETURN
C FIND RANGE OF SUNI
N1 = N-PTS/(RL(1)*2) + 1
N2 = NPTS/(RL(2)*2) +1
N3 = NPTS/(RL(3)* 2) + 1
DO 20 NX=-N1, N1
DO 20 NY =-N2, N2
DO 20 NZ =-N3, N3

948 J. Acoust.Soc.Am., Vol. 65, No. 4, April 1979 J. AllenandO. Berklev:Methodfor simulating
small-room
acoustics 948
C GET EIGHT IMAGE LOCATIONS FOR MODE it NR
CALL LTHIMAGE (R, R0, RL, NR, DELP)
I0=0

DO 10 L=0,1
DO 10 J =0, 1
DO 10 K=0, 1
I0=I0+l
C MAKE DELAY AN INTEGER
ID =DELP (I0) +.5
FDM1 =ID
ID=ID+i
IF(ID.GT.NPTS)GO TO 10
C PUT IN LOSS FACTOR ONCE FOR EACH WALL REFLECTION

1 *BETA(2,1)**IABS(NX)

GID
=BETA(1,1)**IABS
2 *BETA(I,2)** IABS(NY-J)
3 *BETA(2,2)** IABS(NY) NOTE CONTINUATION
4 * BETA (1,3'* IABS (NZ-K) LINES

5 *BETA(2,3)**IABS(NZ)
6 /FDM1
C CHECK FOR FLOATING POINT UNDERFLOW HERE;
C IF UNDER FLOW, SKIP NEXT LINE
HT (ID) = HT (ID) + GID
10 CONTINUE
20 CONTINUE
C IMPULSE RESP HAS BEEN COMPUTED
C FILTER WITH HI PASSFILT OF 1%OF SAMPLING FREQ (I.E'. 100 HZ)
C IF THIS STEP IS NOT DESIRED, RETURN HERE
W = 2.* 4.*ATAN(1 .)* 100.
T = 1E-4

R1 = EXP (-W* T)
R2 =R1
B1 = 2.* Ri* COS (W* T)
B2 = -RI*R1
A1 =-(1. +R2)
A2 =R2
Yi=0
Y2=0
Y0 =0
C FILTER HT
DO 40 I=1, N-PTS
X0 =HT(I)
HT(I) =Y0 +AI*YI+A2*Y2
Y2 =Y1
Y1 =Y0
Y0 =Bi*Y1 +B2*Y2 +X0
40 CONTINUE
RETURN
END

C PGM: LTHIMAGE
PGM TO COMPUTE EIGHT IMAGES OF A POINT IN BOX

SUBROUTINE LTHIMAGE(DR, DR0, RL, NR, DELP)

DR IS VECTOR RADIUS TO RECEIVER IN SAMPLE PERIODS

DR0 IS VECTOR RADIUS TO SOURCE IN SAMPLE PERIODS
RL IS VECTOR OF BOX DIMENSIONS IN SAMPLE PERIODS
Nit IS VECTOR OF MEAN IMAGE NUMBER
DELP IS VECTOR OF EIGHT SOURCE TO IMAGE
DISTANCES IN SAMPLE PERIODS

DIMENSION R2L(3), RL(3), NR(3),DELP(8)

DIMENSION DR0(3), DR(3), RP(3,8)
C LOOP OVER ALL SIGN PERMUTATIONS AND COMPUTE H + / -- H0
I0= 1
DO 10 L =-1,1,2
DO 10 J =--1,1,2
DO 10 K=-1,1,2

949 J. Acoust.Soc. Am., Vol. 65, No. 4, April 1979 J. Allen and D. Berkle¾:Method for simulatingsmall-roomacoustics 949
C NEAREST IMAGE IS L=J=K=--I
RP (1, I0) = DR(1) + L'DR0(1)
RP(2,I0) =DR(2) +J'DR0(2)
RP(3, I0) =DR(3) +K'DR0(3)
IO=IO+l
10 CONTINUE
C ADD IN MEAN RADIUS TO EIGHT VECTORS TO GET TOTAL DELAY
R2L (1) = 2.* RL (1)* NR(1)
R2L (2) = 2.*RL (2)* NR(2)
R2L (3) = 2.*RL(3)*NR(3)
DO 20 I=1,8
DE LSQ: 0
DO 25 J=l,3
R1 = R2L (J)-RP(J, I)
DELSQ =DELSQ +Rl**2
25 CONTINUE
DELP(I) = SQRT(DELSQ)
20 CONTINUE
RETURN
END

IBarbaraMcDermottandJontAllen, "Perceptual
Factorsof an Impedance Boundary," J. Acoust. Soc. Am. 59, 780-785
Small Room Reverberation," J. Acoust. Soc. Am. 60, S9 (A) (1976).
(1976). lSR.J. Donato, '•SphericalWaveReflectionsfrom a Boundary
2JontB. Allen andB. J. McDermott,"A NewMethodfor Mea- of Reactive Impedance using a Modification of Cagniard's
suring Perception of Room Reverberation" (unpublished). Methods," J. Acoust. Soc. Am. 60, 999-1002 (1976).
3C.F. Eyring, "Reverberation
Time {n'Dead'Rooms,"J. 16JontB. Allen, D. A. Berkley, andJ. Bsuert, "Multimicro-
Acoust. Soc. Am. 1, 217-241 (1930). phone sig•l-processor technique to remove room reverbera-
4D.Mlntzer, "TransientSoundsin Rooms,"J. Acoust.Soc. tion from SpeechSignals," J. Acoust. Soc. Am. 62, 912-915
Am. 22, 341-352 (1950). (1977).
•R. H. Bolt, P. E. Doak, andP. J. Westerveit, "Pulse Sta- l?S. T. NeeIcyandJont B. Allen, "Invertibfiity of a RoomIm-
tistics Analysis of Room Acoustics," J. Acoust. SOc. Am. pulse Response," J. Acoust. Soc. Am. (in press).
22, 328-339 (1950). 18p.M. Morse andK. U. Ingard, TheoreticalAcoustics
6j. M. Betman,"Behavior
of Sound
in a Bounded
Space,"
J. (McGraw-Hill, New York, 1968), Sec. 9.5.
Acoust. Soc. Am. 57, 1275-1291 (1975).
lSWithoutthe high-passfilter, the Fourier transformofp (t)
?E.K. Dunens
andR. F. Lambert."Impulsive
Sound
Response shows a large energy spike at zero frequency as the absorp-
Statistics," J. Acoust. Soc. Am. 61, 1524-1532 (1977).
tion a goes to zero. This is in part a direct result of the no-
6J.R. Power, "Measurement
of Absorption
in Roomswith
pressure-release boundary conditions we have assumed and
SoundAbsorbingCeilings," J. Acoust. Soc.Am. 10, 98-101
(1938). is further complicated by the assumption of an accelerative
source, which is nonphysical for very low frequencies (less
9C.G. Mayo,'•tanding
WavePatternsin Studio
Acoustics," than 50 Hz).
Acustica 2, 49-64 (1951).
20j. j. Jetzt, "Critical DistanceMeasurementof Roomsfrom
1øB.M. GibbsandD. K. Jones,"A SimpleImageMethodfor
the SoundEnergy Spectral Response," J. Acoust. Soc. Am.
Calculating the Distribution of Sound Pressure Levels within
(to be published).
an Enclosure," Acustica 26, 24-32 (1972).
2tJontB. Allen, '•ASTFILT-An FFT BasedFiltering Pro-
tlF. Santon,"NumericalPredictionof Echograms
andtheIn-
graxn, "in IEEE Press bookon Programs fo• Digital Signal Pro-
telligibfiity of Speechin Rooms,"J. Acoust. Soc.Am. 59, cessing (IEEE Press, New York, 1979).
1399-1405 (1976).
zzM.R. Schroeder,"NewMethodof Measuring
Reverberation
2By"approximate"
wemeantheblurringoftheimageasa re- Time," J. Acoust. Soc. Am. 37, 409-412 (1965).
suit of the finite impedancewall. Variations in wall imped- 2SAgreement
is best for beta the same on all surfacesand
ance as a functionof frequencyor other frequencydispersion variations appear when a pair of opposingwalls are signifi-
effectswill also make the imagesfrequencydependent.In cantly different in reflectivity th•n all other surfaces, (as is
practice the two effects usually appear simultaneously. the case for Figs. 2 and 3).
t3A.R. Wentzel,"Propagation
of WavesAlonganImpedance 24D.A. Berkley,"NormalListeners
In TypicalRooms:
Rever_
Boundary,"J. Acoust. Soc.Am. 55, 956-963 (1974). berationPerception,SimulationandReduction"(unpublished).
14S.Thomasson,
"Reflection
of Wavesfroma PointSource
by

950 J.Acoust.
Soc.Am.,Vol.65,No.4, April1979 J.AllenandD. Berkley:
Method
forsimulating
small-room
acoustics 950

View publication stats