Schroeder Quadratic Residue Diffusors
Schroeder Quadratic Residue Diffusors
Schroeder Quadratic Residue Diffusors
DrittesPhysikalisches lnstitut, Universitdt G6ttingen, Germany and Bell Laboratories, Murray Hill, New Jersey 07974 (Received2 October 1978)
Ceiling shapes concert for hallsareproposed on thebasis priorextensive that, of subjective evaluations, should result highlistener in preferencethehall's of acpistoca; responsemusic. to These shapes based are on the premise as littleas possible the earlysound that of energy should arriveat a listener's in the ears "median"plane(the verticalsymmetry planethroughthe listener). While this goal is inherently approachedold-style, in high-and-narrow its realization modern, halls, in low-ceiling requires halls special ceiling shapes surface and structures keepearly,median-plane to sound awayfromthelistener's ears.
PACS numbers:43.55.Br, 43.55.Fw, 43.66.Pn
I. BINAURAL
DISSIMILARITY
planeas opposed the (vertical) medianplane. Such to early lateral reflections lead to agreaterbinaural dissimiarity,
andtherefore higher listener preference. Others whohave
discussed the effects of early lateral reflections include
Specifically, for concert hails and seat positions with highly similar ear signals, the subjective preference was low. Conversely, for more dissimilar ear signals
the preference was high. In other words, binaural dis-
importance of early lateral reflections has been obtained by the author and his collaborators
gen? 'e While in the study of real halls different physical parameters come in a "predetermined mix," which
can only be disentangled statistically by appropriate subjective scaling methods, the technique of digital
similarity and subjective preference are positively correlated. In fact, binaural dissimilarity was found to be at least as strongly correlated with preference as any
other objective parameter--including
time.
reverberation
The musical program material for these studies was from the classical and romantic repertoire. However, considering the fundamental invariance of all musical perception--as opposed to speech perception--one may expect corresponding results for other musical styles
as well.
Accordingly, binaural impulse responses from real halls were fed into a computer and digitally modified by adding (or deleting) lateral reflections. When the effect of these modifications is expressed as binaural simil-
Binaural similarity (called "interaural coherence" and abbreviated"C" in Ref. 1) is defined as the peak
value of the correlation function of the first 80 ms of
naural similarity?
the impulse responses within an interaural delay range of 1 ms. Binaural dissimilarity is defined as the negative of binaural similarity.
The results imply that sound arriving in the median
score was obtained for the highest value of binaural similarity. Interestingly, low preference was also
plane of a listener is detrimental to subjective preference because such median-plane sounds lead to identi-
cal soundpressure waves at the two ears thereby decreasing binaural dissimilarity. (An explanation for the
How then can we design concert hall ceilings that avoid direct (median-plane) reflections into the audience area? Absorption is vetbotch because we need the energy from the ceiling for reverberation. However, we
can shape the ceiling to reflect most of the soundenergy to the side walls, whenceit will reach the listener, directly or indirectly, from lateral directions.
waswith thewidth of the hall; the narrower the hall, the higherthepreferenceandvice versa. Narrow halls, of course,havemore numerous more powerfulearly latand
eral reflections that arrive at the ears near the horizontal
958
0001-4966/79/040958-06500.80
rays from one of its focal lines as if they were coming from the other focal line. Thus, if the outer focal line
of the ceiling's hyperbolic cross section is located in the stage area at the height of the musical instruments, the reflected soundrays will seem to come from a point
only slightly above the ceiling and will spread out accordingly with much more energy going laterally than
for a plane ceiling.
r, =exp(i2rrs,/N) ,
has constant magnitude. In fact,
(2)
[R.I2=
expi-iZnm/v)l'--1/v.
(3)
Of the several proofs of this relationship, a particularly simple one is based on the fact that the autocorrela-
tion sequenceof r. is zero except for a zero shift (modulo N). = And this, in turn, is a consequenceof the fact that the s, form a complete deference set, i.e.,
that no matter what interval k one considers, the pair-
CEILINGS
wise differencess,. k- s, (rood3/) generateall the numbers between 0 and 3/- 1 exactly once per period. Thus, for k = 1, for example, the differences generated are:
tion, scattering a simple ray into ten or more "raylets" with roughty equal energies over a wide angular distribution. Surfaces with such highly diffuse reflections including experimental results were described
is flat.
accord-
However, becausethe of quarter wavelength requirement, the sound diffusing properties of the surface de-
penduponthe wavelengthof the incident sound. In practice it is foundthat gooddiffusion is obtainedin a range
of wavelengths half an octave below and above the "design wavelength." For an incident wave of half the de-
sign wavelength, the wells are, of course, half a wavelength deep, resulting in near specular reflection by the surface. (However, at one-third the designwavelength, the surface is a gooddiffusor again).
IV. QUADRATIC-RESIDUE DIFFUSORS
with a local impedance g(x) which varies only along one of its dimensions (x) in a periodic fashion with period Nw (see Fig. 1). Z(x) is constant in the direction orthogonal to x. Thus, we can treat the diffraction problem
as a two-dimensional problem, the two dimensions being the x direction and the direction normal to the reflecting surface. By definition:
structures that give excellent sounddiffusionover larger wavelengthranges. In this search, we discovered, by computersimulation, that surfacesbasedon m-ary maximum-length sequences are capableof gooddifz2 fusion over larger bandwidths--presumably because suchsurfaces havewells of several different depths.
This led us to still another kind of mathematical se-
Z(x) :p(x)/o(x) ,
where p(x) is the soundpressure and v(x) is the component of the particle velocity normal to the reflecting surface (the normal being directed into the surface).
(4)
quences, the quadratic-residue sequences elementary of number theory, investigatedby A.M. Legendre and
C. F. Gauss. These sequences are defined as follows
p(,,) =
(5)
(6)
(7)
CEILING
dn,d(x)
NW
LENGTHWISE
FIG. 1. Lateral cross section through diffusely reflecting ceiling based on quadratic residue sequence with N = 17.
(CROSS-DIMENSION
OF HALL)
dividual wells. They are crucial for gooddiffusion. particularly for obliquely
incident sound.
FLOOR--
relation for k, the "spatial frequencies"in the x direction, is the same as that in Eel. 18, Eqs. (2a) and (2b),
except for a change in the notations.)
] p,(x):[z(x) oc]/[z(x)pc], +
02)
v(x)= (iX/2vpc)Op(x)/on .
(8)
Here pc is the characteristic impedance of the medium (air) facing the reflecting surface.
With the standard wave equationfor soundpropagation in two dimensions, one can easily execute the operation ap(x)/an, yielding
i fNw[z(x)_]
Nw
(13)
Next we make the reasonable assumption that the local impedance of a surface with wells such as shown in Fig. I ,has a local impedance
v(x) i(cosa, a,[1 (ksX/2r)2]'/2)/pc. (9) = For values of the summation index s in the range
(10)
[ 1
mo
[-
Nw ]
(14)
the squareroot in Eq. (9) will be real andwill equal the cosineof the scattering angle: cosa. This range
of s corresponds to propagating reflected waves. For s outside the above range, the scattered waves will be evanescentwith zero energy flux. Both types of waves
are important in an exact solution for the unknown scat-
If the d(x) axe constant over a width w that is small compared to the wavelength and then change abruptly to a new value (see Fig. 1) given by the quadratic-resi-
tered amplitude a s.
d. (x/2:V)s.,
with the result, from Eqs. (2) and (3),
(15)
I,,,1 const.-=-I/N,
i.e., under the simplifying assumptions made, the amp-
4 ,
where ' indicates summation s is rethe that index
stricted to propagating waves.
960 J. Acoust. Soc.Am., Vol. 65, No. 4, April 1979
litude of each scattered wave is the same--in a sense, the quadratic-residue surface is an optimum diffusor.
If one desires to go beyond this first-order imation, one has to solve the equation approx'
960
0o
0o
30 30 o
-90o
FIG. 2. Scatterdiagramof the diffusor shown Fig. 1 for in the designfrequency =11.5 kHz. Figures 2-6 are for a f0
normally incident sound wave.
z(x)v(x) :p(x),
rave or more below the frequencycorresponding X to o. However, for wavelengths shorter than ko, the ceiling
remains a good scatterer up to a limit given by prime
for the %, withp(x) as in Eqs. (5)-(7) andv(x) as in Eq. (9). With Z(x) periodic, this is accomplished a by
Fourier transformation, which yields an infinite set of
linear equations. We have found that by restricting this set to 50-100 terms (for N = 17), stable solutionsfor
the as canbe obtained sufficientlyaccuratecompuon ters. All subsequent results were obtainedby inverting
matrices of size 99 x 99.
For the surfaceshown Fig. I (N= 17,w= 0.137ho, in where is the "design"wavelength) vertically inand
cident sound, we obtained the following values for the
five scattering amplitudes:
ever, the energy fluxes, taking the projecting cosines into account, showremarkable uniformity:
: (r=l.7dB,
Xo/2:(y=3.3 dB,
shownin Fig. 3 (for 22.9 kHz), Fig. 4 (for 34.4 kHz), and Fig. 5 (for 45.9 kHz). As canbe seen, there is still gooddiffusion even at the highest frequency. In fact,
the higher the frequency the more scattering angles appear, in accordance with Eq. (10). For intermediate frequencies, i.e., frequencies that
described most effectively by the standard deviation in decibels of the energy fluxes. For the abovecase, (r= 1.7 dB. Thus, while the scattering amplitudesare not uniform, the subjectively important standard deviation of the energy fluxes expressed in decibels is
0o
residue surfaces at wavelengthsother than the design wavelength? For longer wavelengths, the ceiling will look more and more like a smooth mirror, and we may
not expect good scattering for frequencies half an oc961 J. Acoust. Soc.Am., Vol. 66, No. 4, April 1979
-90
FIG. 4. Scatter diagram of the dfffusor shown in Fig. I for f= 3f0= 34.4 kHz.
961
0a
FIG. 7. Scatter diagram for diffusor shown in Fig. 1 for an obliquely (c, -55 ) incident sound wave for j =22.9 kHz. The
90
scattered energy covers the entire range from -90 to +90 , but is somewhat less uniformly distributed as for normal incidence at the same frequency (of. Fig. 3).
are not integer multiples of the design frequency, similarly good results are obtained. For ten randomly selected frequencies, in the first and second octaves above the design frequencies the following average standard
deviations were found computationally:
V. TWO-DIMENSIONAL
SCATTERERS
So far one-dimensional sound scattering has been considered. In other words, if the wells on the ceiling run lengthwise, sound waves from the stage are diffused laterally from the ceiling, but are reflected mirror-like in the front-back dimension.
If we desire also longitudinal diffusion, the ceiling surface needs structuring also in the longitudinal direction. This is easily realized by replacing the quadratic
residue sequence of Eq. (1), by a ho-dimensional quadratic residue array of reflection coefficients.
r, =exp[i2r(k2 + 12)/N],
distance vz in the front-back
Such two-dimensional
(16)
direction.
were constructed on
the quadratic residue diffusors? For four different diffusors with rmdomly selected well depths, the following standard deviations of the energy fluxes were
obtained
a model scale. Measurements confirmed the expected excellent diffusion over the entire solid angie. For an illustration of such a diffusor, see Fig. 8.
FIG. 6. Scatter diagram of the dfffusor shown in Fig. 1 for a nonintegral frequency f =32.4 kHz.
962 J. Acoust.Soc. Am., Vol. 65, No. 4, April 1979
90 - 50
9o
FICa. 8. Photograph of scale model of two-dimensional dfffusor based on quadratic residue array, for efficient diffusion over
the entire solid angle.
962
If "back scattering" (backwardfrom the ceiling to the stage) is desired even at the highest frequency the longitudinal widths w, shouldbe made aquarter-wavelength
rather than a half-wavelength.
If esthetically preferred,
The two ceiling designs for maximizing early lateral reflections described heretoone based on geometrical
principles, the other on sounddiffraction--can be profitably combined. The diffraction principle allows either one or twodimensional diffusion--with different frequency ranges,
if desired.
The amount of flexibility inherent in the proposals made here should be sufficient to achieve the preponderance of lateral reflections necessary for the high listener preference found in concert halls with large
binaural dissimilarity.
I thank R. C-erlach, D. Gottlob, H. Henze, and A. Steingrube for stimulating discussions and for performing some of the measurements and numerical computations.
tM. R. Schroeder, "Computer Models Concert for Hall Acoustics," Am. J. Phys.41/4, 461-71 (Aprll 1973). isM. R. Schroeder, "Diffuse Sound Reflections Maximumby
Length Sequences," Acoust. SUc.Am. 57, 149-150 (1975). J.
Martmnlfolgendiffusoren einem Modellhailraum," Fortech. in Altoslik, DAGA '76, Heidelberg, VDI-Verlag, Duesseldorf
(1976) p. 255-258.
2p. Damaske, "SubJektive Untersuchung Schallfeldern*" you Acustica19, 199-213 (1967/68). 3W. Burgtorfand B. Wagenet, "Maskingby Subjectively Diffuse Sound Fields," Acustica19, 72 (1967/68). 4H. Kuttruff, Room Acoustics (Applisd Science,London,1973),
pp. 192--199.
Town Hall. J. Acoust. Soc. Am. 63, S36 (A) (1978). See also
963
963