Вы находитесь на странице: 1из 13

Denition and measurement of

random-incidence scattering coecients


Michael Vorla nder
a,
*, Eckard Mommertz
b
a
Institute of Technical Acoustics, Rheinisch-Westfalische Technische Hochschule, Templergraben 55,
D-52056 Aachen, Germany
b
Muller-BBM GmbH, Robert-Koch-Str. 11, D-82152 Planegg, Germany
Received 11 January 1999; received in revised form and accepted 25 June 1999
Abstract
This paper describes the denition and two methods for measuring scattering coecients of
rough surfaces. In both cases, impulse responses are determined for various orientations of a
sample surface. After phase-locked superposition of a sucient number of impulse responses,
the coherent reected sound energy is obtained. It is identical with the zero-order lobe of the
reection pattern the specularly reected part. Considering the totally reected sound
energy, the scattering coecients can be calculated. The results of the reverberation chamber
measurements are scattering coecients for random sound incidence. Both methods are dis-
cussed and experimental results, obtained by scale model measurements, are shown. # 2000
Elsevier Science Ltd. All rights reserved.
1. Introduction
Some problems in acoustics require knowledge about scattering of sound at cor-
rugated surfaces. This is true, for instance, for characterisation of the seaoor, of
road barriers or in room acoustics. Particularly, in computer simulation of sound
propagation in rooms, realistic results are obtained when not only specular reections
are considered but also diuse reections. This is necessary since generally the irregu-
larities of wall surfaces are neither small nor large compared to the wavelengths.
Traditionally, the scattering properties are expressed in terms of the directivity
pattern. These can be measured or calculated for distinct angles of incidence and
frequencies (e.g. [14]). However, only in a few cases the detailed directivity pattern of
Applied Acoustics 60 (2000) 187199
www.elsevier.com/locate/apacoust
0003-682X/00/$ - see front matter # 2000 Elsevier Science Ltd. All rights reserved.
PI I : S0003- 682X( 99) 00056- 0
* Corresponding author. Tel.: +49-241-8079-85/86; fax: +49-241-8888-214.
E-mail address: mvo@akustik.rwth-aachen.de (M. Vorla nder).
the scattered sound is of major interest. Instead, an adequate quantity for describing
rough surfaces is the scattering coecient o. It is simply dened as the ratio of the
non-specularly reected sound energy to the totally reected energy (Fig. 1) and
does not include any information about the directivity of the scattered energy.
However, although generally not physically exact, in room acoustical prediction
methods the directional energy distribution may be expressed in terms of Lambert's
cosine law [5], if at least o is known.
The scattering coecient o generally depends on the frequency and on the angle of
sound incidence. However, analogue to the random-incidence absorption coecient
obtained in reverberation rooms, an angular average of the scattering coecient, i.e.
the random-incidence scattering coecient, can be dened as well.
The scattering coecients can be calculated from the directivity pattern (see also
[6], in this issue). However, in this paper two measurement methods are described
which allow determination of the scattering coecients directly, as presented earlier
in [7].
2. Fundamentals
The procedures are based on the idea presented already in [8,9]. The general
principle of the method can best be explained by looking at the eect of reection
and scattering in the time domain. In Fig. 2 three bandpass ltered pulses are
shown, which were reected from a corrugated surface for dierent orientations of
the sample.
Obviously the initial parts of the reections are highly correlated. In contrast, the
later parts (``tails'') are not in phase and depend strongly on the specic orientation.
This observation makes sense, because the rst part of each reection took the path
with the shortest delay (Fermat's principle, see Fig. 3). This coherent part is identical
with the specular component of the reection which can also be interpreted as the
Fig. 1. Scattering from rough surfaces.
188 M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199
zero-order lobe of the directional pattern. The energy in the ``tail'' of the reected
pulse contains thus the scattered part.
The energies (normalised with respect to a reection from a rigid reference plane)
contained in the impulses in Fig. 2 are expressed in terms of (see also Fig. 1):
E
spe
= 1 o ( ) 1 o ( ) = 1 a ( ). E
totl
= 1 o ( ) (I)
The coecient a may be called ``specular absorption coecient''. From these equa-
tions the scattering coecient o can be determined by:
o =
a o
1 o
= 1
E
spe
E
totl
(P)
Fig. 2. Exemplary reected pulses (10 kHz, 1/3 octave band) obtained for dierent sample orientations.
Fig. 3. Reection paths: (F F F) specular according to Fermat's principle and () scattered. R 0
1
. c
1
. 0
2
. c
2

indicates that the reection amplitude depends on the two incident angles and the two reection angles in
the local polar coordinate system at the reection point.
M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199 189
As can be easily seen, the ``specular absorption coecient'' directly corresponds to
the scattering coecient if o % 0.
The important idea of the approach is to extract the specular component E
spe
from the reected pulses. This is done by phase-locked averaging of the pulses
obtained for dierent sample orientations: while the specular components add up in
phase the incoherent scattered sound interferes partly destructive. This principle is
applied and further explained in the two measurement methods described in the
following.
3. Free-eld method
To demonstrate the method, simple model structures consisting of randomly dis-
tributed, parallel wooden hemi-cylinders and rectangular battens were used as test
samples. The measurement set-up is illustrated in Fig. 4.
Instead of using band-limited pulses as described above, the measurements were
performed with broadband MLS excitation using a PC-based measurement system.
The evaluation was done in the frequency domain after Fourier transform which is
equivalent to algorithms dened in the time domain.
In the rst step, the complex reection factors R
i
f are determined (see [10]) for
dierent orientations of the sample. Each reection factor R
i
f may be expressed in
terms of a specular and a diuse component:
R
i
f ( ) = R
spe
f ( ) S
i
f ( ). (Q)
The specular complex reection factor is obtained by coherent addition of n com-
plex reection coecients under variation of the orientation of the sample c
i
. While
the energy of the specular component jR
spe
f j
2
adds up coherently proportional to
n
2
, the scattering components are incoherent. Their energy increases proportional to
n. With suciently large number of averages the incoherent part becomes su-
ciently small. We obtain the specular reection factor:
Fig. 4. Experimental arrangement.
190 M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199
R
spe
f ( )
1
n

n
i=1
R
i
f ( ) for n >> 1 (R)
The specular absorption coecient a is calculated by the equation:
a f ( ) = 1 [R
spe
f ( )[
2
(S)
Provided the energy losses can be neglected, the result of Eq. (5) is directly iden-
tical with the scattering coecient o. Otherwise the absorption coecient must be
determined separately. In rst-order approximation this is achieved by quadratic
summation of the individual (angle-dependent) results:
o f ( ) - 1
1
n

n
i=1
[R
i
f ( )[
2
. (T)
It should be noted that in some cases it is necessary to perform additional measure-
ments with the microphone placed at a larger distance from the test object to avoid
near eld eects. This involved a check if the far-eld impulse response contained no
components signicantly dierent from those in the near-eld situation.
In principle the method of averaging is possible for all angles in the complete
hemisphere above the sample. But it is also possible to average just over the azimuth
angles. In this case, since the samples are rotated between two measurements by c,
as indicated in Fig. 4, the scattering coecient o 0 averaged over the horizontal
angles c for one azimuth angle 0 is obtained. By applying Paris' formula [5] to the
absorption coecients collected for dierent azimuth angles 0 it is possible to obtain
the random-incidence scattering coecient o
s
.
Fig. 5 shows the results obtained for the two test structures in a scale model. To
illustrate the size of the eects, it should be noted that the wavelength at 10 kHz is
approximately equal to the average structure period d. Accordingly dl equals 1
at 10 kHz. The measurements were performed at 0 30

. 45

. 60

. 75

and c
0

90

with c 10

. Slightly more measurements at dierent angles are useful to


increase the accuracy.
However, the absorption coecient of materials for random sound incidence is
not determined by tedious angle-dependent measurements and averaging, but
directly in the diuse eld of a reverberation chamber according to ISO 354. Exactly
the same approach can be applied in measurement of the scattering coecients, too,
as described in the next chapter.
4. Reverberation chamber method
4.1. Description of the method
Measurements of absorption coecients in the reverberation room according to
ISO 354 are performed by measurement of the reverberation time with (T
2
) and
without (T
1
) the test sample. By using the relation
M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199 191
o

= 55.3
V
cS
1
T
2

1
T
1

(U)
one obtains the random-incidence absorption coecient o

of the test sample. Here,


V is the room volume, c the speed of sound and S the area of the test sample.
The basic idea of measuring scattering coecients in the reverberation room is the
same as described in the previous chapter. However, the turntable with the sample is
now placed in a scale model reverberation room (Fig. 6). Source and receiver are
xed, while the test sample is turned during the measurement stepwise by c. If the
sample reects purely specularly (and if the sample area is circular), the measure-
ment results are perfectly correlated. This is true for any shape and wall construction
of the reverberation chamber. Provided the sample introduces diuse reections,
the room impulse response shows slight dierences in the ne structure (Fig. 7).
These dierences become larger at larger delays, since late reections suered scat-
tering at the test sample more often. The average energy decay, however, remains
unchanged.
Once more, it is the aim to suppress the sound components scattered from the
sample by phase-locked averaging the room impulse responses obtained for dierent
angles c
i
. Assuming statistical independence between specular and scattered com-
ponents, it can be shown (see Appendix) that after addition of n room impulse
responses, the energy decay E t (modelled by an exponential decay) can be
approximately expressed in terms of (see Appendix):
Fig. 5. Measurement result for rectangular and hemi-cylindrical battens placed on a reecting surface.
192 M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199
Fig. 6. Scale model reverberation room (V=1 m
3
) and sample placed on turntable.
Fig. 7. Initial part and late part of room impulse responses. Upper diagrams: with xed sample, lower
diagrams with rotating sample.
M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199 193
E t ( ) - n 1 ( )e
(cS4V) ln 1a ( )t
e
(cS4V) ln 1o ( )t
.
a =
1
S
S S
smple

o
0
S
mple
a
mple

(V)
with o
0
denoting the mean absorption coecient of the empty room. Hence, the
latter exponential function can be neglected with suciently large n as illustrated in
Fig. 8. Accordingly, by evaluation of the early part of E t we obtain a reduced
reverberation time T
2
(see also Fig. 12 below).
Following the procedures described in ISO 354, the absorption coecients o
s
and
a
s
can be determined. Finally, the random-incidence scattering coecient o

is cal-
culated according to Eq. (2).
4.2. Discussion of exemplary results
Examples of results obtained in the scale model reverberation chamber in com-
parison with the free-eld method are shown in Fig. 9. At low and mid frequencies
the agreement between results from the free-eld method and the reverberation
room method is very good indeed. At high frequencies some deviations are
observed.
The deviations at high frequencies can be explained by the fact that the sample
area was not circular but quadratic. This inuence is larger with the reverberation
room method, since the distance of the microphones to the sample was larger.
However, the eect of the base surface size and shape must be studied more inten-
sively in future.
In further tests with other exemplary samples the circular arrangement was used.
Fig. 10 shows results of wooden hemispheres distributed randomly on a wooden
Fig. 8. Impulse responses measured in the reverberation room (10 kHz, 1/3 octave band).
194 M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199
plate at various densities. Fig. 11 shows scattering coecients as well as the
absorption coecient of a scale modelled Schroeder diuser (circular sample). This
kind of material is particularly used in studio constructions for scattering purposes.
(The sample was granted by RPG Diusors Inc. for the measurement series.)
4.3. Required number of averages
In the following it is discussed, how the number of averages aects the resulting
``integrated impulse responses''. For this purpose impulse responses were recorded
at subsequent angular steps of 10

. The systematic inuence of averaging was


investigated by using n=1, 4, 8, 18 and 36 of the total 36 separately obtained
impulse responses.
The curves in Fig. 12 clearly show dierent decays contained in the average curve.
The late decay constant is related to the xed sample (and thus to the ``regular''
Fig. 9. Scattering coecients of the samples shown in Fig. 4, measured in the reverberation chamber (&)
and in the free eld (^).
Fig. 10. Scattering coecients of the circular plate covered with wooden hemispheres measured in the
reverberation chamber. Parameter is the density of the scatterer.
M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199 195
average absorption coecient). The early decay, however, is much steeper. The
``bended'' decays indicate the superposition of two independent decays, similar to
that observed in coupled rooms. The initial level of the late decay decreases by 3 dB
per doubling the averages, and the time corresponding to the decay intersection
increases. Since we are interested in the rate of the early decay, it is reasonable to use
the reverberation time T
10
or T
20
(decay from 5 to 15 or 25 dB, respectively).
Furthermore, in Fig. 12 the result with continuously turned sample is shown.
During one complete revolution, 94 periods of the test signal were radiated con-
tinuously, detected and averaged. This corresponds to a total measurement time of
80 s. Although the measurement technique (MLS) assumes time-invariance, mea-
surements of this kind can be performed since only the time-invariant components
of the room impulse response are to be determined. Additionally, this method is
Fig. 11. Scattering and absorption coecients of a RPG-Diusor.
Fig. 12. Integrated impulse responses (Schroeder plots), obtained after phase-locked averaging of n room
impulse responses (10 kHz 1/3 octave band). Test sample: randomly distributed hemispheres (=4 cm,
density 50%) on a rigid circular plate (=60 cm).
196 M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199
advantageous because less data must be stored in the computer memory and the
measurement is performed much faster. The latter criterion is important in view of
time variances caused by temperature variations or similar eects [11].
5. Conclusions
It can be stated that the reverberation method is well suited for estimating a gen-
eral measure for the scattering properties of reective surfaces the scattering
coecient for random sound incidence. As has been mentioned, scattering coe-
cients can be used for instance as input data for room acoustical simulations. Fur-
thermore, they can characterise the scattering properties of architectural surfaces
which are an important factor in room acoustical design.
Only if deeper knowledge on the directional pattern of the scattered sound is
required, measurements or calculations of the polar response become necessary.
It is worth mentioning that this approach is one of the favourites for being stan-
dardised in ISO/TC43/SC2. A working group was mandated to develop a denition
and a measurement method for sound scattering.
Acknowledgements
Y.W. Lam is acknowledged for inviting us to publish a paper on our work. This
work was in part supported by the Deutsche Forschungsgemeinschaft DFG.
Appendix: Expectation value of the average room impulse response
In order to explain the structure of the decay curve shown in Fig. 12, energy
impulse responses after coherent addition of n bandpass ltered impulse responses
are considered:

n
i=1
h
i
t ( )

2
= nr
spe
t ( )

n
i=1
s
i
t ( )

2
(eI)
Here, r
spe
t corresponds to the coherent component in each measurement. Pro-
vided the scattered components of the sample s
i
t are statistically independent, one
yields with Eq. (A1):

n
i=1
h
i
t ( )

2
- nr
spe
t ( )

n
i=1
s
i
t ( )

2
= n
2
r
spe
t ( )

2
n - s t ( )

2
> (eP)
with
M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199 197
- s t ( )

2
>=
1
n

n
i=1
s
i
t ( )

2
-
1
n

n
i=1
s
i
t ( )

2
. (eQ)
If jr
spe
t j
2
is modelled by an exponential decay, as usually done in statistical
reverberation theory, the coherent component is
E
spe
t ( ) = E
0
e
13.8tT
2
(eR)
while the scattered energy E
st
t is identical with the dierence between the totally
and the coherently reected energy:
E
st
t ( ) = E
totl
t ( ) E
spe
t ( ) = E
0
e
13.8tT
1
E
0
e
13.8tT
2
. (eS)
Using Eqs. (A2), (A4) and (A5) we obtain the average energytime curve after
addition of n room impulse responses:

n
i=1
h
i
t ( )

2
=

nE
0
n 1 ( )e
13.8tT
2
e
13.8tT
1

. (eT)
It is indeed the superposition of two decay curves with dierent initial levels and
decay constants. Of course it must be taken into account that an increase of the
number of averages is corresponding to smaller angular steps. These, however, must
be suciently large in comparison of the wavelengths with the travelled distance of
surface elements, to ensure decorrelation of the scattered components from mea-
surement to measurement.
References
[1] Meyer E, Bohn L. Schallreexion an Fla chen mit periodischer Struktur. Akustische Beihefte zur
Acustica 1952;3:AB195207.
[2] D'Antonio PJ, Konnert JH, Kovitz PS. Experimental measurements of directional scattering prop-
erties of architectural acoustic surfaces. Proceedings W.C. Sabine Centennial Symposium, Cam-
bridge, MA 1994. p. 1414.
[3] Schroeder MR. Phase gratings with suppressed specular reection. Acustica 1995;81:35469.
[4] Kleiner M, Gustafsson H, Backman J. Measurement of directional scattering coecients using near-
eld acoustic holography and spatial transformation of sound elds. J Audio Eng Soc 1997;45:331
46.
[5] Kuttru H. Room acoustics. 3rd ed. London: Elsever Applied Science, 1991.
[6] Mommertz E. Determination of scattering coecients from the reection directivity of architectural
surfaces. Applied Acoustics 2000;60(2):2013.
[7] Mommertz E, Vorla nder M. Measurement of scattering coecients of surfaces in the reverberation
chamber and in the free eld. Proceedings 15th International Congress on Acoustics, Trondheim,
1995. p. 57790.
[8] Vorla nder M. Messung der Schallstreuung an rauhen Fla chen. Fortschritte der Akustik, DAGA88,
Braunschweig. p. 7614.
198 M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199
[9] Vorla nder M, Schaufelberger T. Messung des Diusita tsgrades rauher Fla chen. Fortschritte der
Akustik DAGA'90, DPG-GmbH, Bad Honnef, 1990. p. 82730.
[10] Mommertz E. Angle dependent in-situ measurements of reection coecients using a subtraction
technique. Applied Acoustics 1995;46:25163.
[11] Vorla nder M, Bietz H. Der Einu von Zeitvarianzen bei Maximalfolgenmessungen. Fortschritte der
Akustik DAGA'95, DPG-GmbH, Bad Honnef, 1995.
M. Vorlander, E. Mommertz / Applied Acoustics 60 (2000) 187199 199

Вам также может понравиться