Bengt-Inge Dalenbäck, Ph.D., CATT
www.catt.se
bid@catt.se

0. Introduction

This is the second part of a two-part on-line paper discussing reverberation time estimation with special emphasis on the effects of diffuse reflection in computerized prediction (CP) in relation to classical Sabine methods. The purpose of this paper is to discuss problems, pitfalls and techniques regarding reverberation time (RT) prediction and gives examples from idealized as well as actual rooms encountered in consulting practice.

Part I addressed:

1. What effect has diffuse reflection on the RT?
2. Will diffuse reflection always affect the RT?
3. What effect has diffuse reflection on Sabine RT estimates?
4. Will not formulas such as Fitzroy and Arau-Puchades solve the problem?

Part II addresses:

1. Does diffuse reflection have to be taken into account?
2. How can diffuse reflection be handled in a CP program?
3. Must diffuse reflection be handled with frequency dependence?
4. How can scattering coefficients be estimated?
5. How will diffuse reflection affect auralization?

1. Does diffuse reflection have to be taken into account?

No not always. As indicated in part I, for mixing room shapes the RT can be well predicted even without diffuse reflection (even if prediction of finer parameters such as C-80 may suffer). However, the central question is then: how can it be known in advance if a room shape is mixing? The answer is that it cannot, at least not just from inspecting the room shape except in some extreme cases where it can be deduced that the shape is not mixing. There is no reason not to include diffusion in a CP program, it will only mean an unnecessary uncertainty and a potentially dramatic overestimation of the RT.

2. How can diffuse reflection be handled in a CP program?

The most common way for CP programs to handle diffuse reflection is by randomizing the direction of reflected rays according to some distribution [Kuttruff]. The procedure is as follows: when a ray is to be reflected from a diffusing surface a random number A in the range 0 to 1.0 is generated. If this number is higher than the scattering coefficient s of the surface the reflection is specular, if it is lower the ray will be diffusely reflected (if the scattering coefficient is 0.30 then 30% of the rays that hit the surfaces will be diffusely reflected and 70% will be specularly reflected). If the value of A was such that the reflection should be diffuse, two new random numbers, B and C, are generated that decide in which direction the ray should be reflected according to the direction distribution function used. 

Figure II-1 Procedure to create diffuse reflection by randomizing ray directions.

The most common diffuse reflection distribution function is that of Lambert (ideal diffuse reflection, often used in optical ray-tracing) [Kuttruff]. Although there are reasons to claim that Lambert is not a good function to use, since it originates from optics where the wavelengths are much smaller than in acoustics, it is a distribution for which a lot of experience has been collected since it has been - and still is - used in most CP programs that handle diffuse reflection. CATT-Acoustic™ employs Lambert as described for two calculation methods: "Audience area mapping" that is based on ray-tracing with fixed-sized receiver spheres and in "Full detailed calculation" that is based on the unique Randomized Tail-corrected Cone-tracing (RTC). However, the RTC uses the described procedure only for reflection orders greater than one. For first and second order specular reflections, the image source model is used while first order diffuse reflection is handled by distributing many small sources over each diffusing surface. To have the highest geometrical accuracy where it is needed, the lower the absorption coefficient and the higher the diffusion coefficient the more surface sources are used. From the actual sound source, vectors are drawn to each diffuse surface source and from each of those to the receivers (taking occlusion into account), figure II-2 illustrates the procedure. 

a) 1st order specular and diffuse reflections

b) Schematic echogram Receiver 1
(direct sound not shown). Specular 
reflection is attenuated by (1-s)

c) 1st order diffuse reflections

d) Schematic echogram Receiver 2
(direct sound not shown)

Figure II-2 Procedure used to calculate first order diffuse reflection in the RTC. a), b) receiver in specular zone, c),d) receiver outside specular zone.

The strength of each diffuse reflection is, according to Lambert, proportional to the cosine of the incidence angle as well and the cosine of the reflection angle (as measured towards the surface normal). However, since there are many surface sources, all with different incidence and reflection angles, the total effect for surfaces of some size is not that of a cosine directivity but rather something that is quite similar to a QRD™ diffuser. This method will, as is indicated in figure II-2 give the smear of reflections in time and many other properties of real diffusers if not in a detailed manner. It would be possible to use distribution functions measured from actual diffusers and that is a likely future development for CP programs.

3. Must diffuse reflection be handled with frequency dependence?

Since diffusion is strongly related to the ratio between surface roughness/size the obvious answer is Yes. Figure II-3 shows a schematic illustration of how the ratio affects diffusion. 
a) wavelength >> D : low diffusion

b) wavelength = D : high diffusion,
complex actual behavior

c) wavelength << D : geometrical mixing 
(effectively acts as diffusion)

Figure II-3 Schematic description of how the ratio between surface roughness and wavelength determines the diffusion (D is a typical size of the roughness). 

This is of course a simplification but it illustrates the necessity to handle diffusion with frequency dependence. In addition, available measurements such as on the RPG site (Product Specifications section, e.g. http://www.rpginc.com/products/qrd734/qrd_dc.htm and http://www.rpginc.com/products/skyline/sky_dc.htm) clearly indicate that a single value for all bands will not suffice. Indeed, for some diffuser designs, the diffusion coefficient may vary more with frequency than its absorption coefficient does, this behavior may also be carried over to the scattering coefficient. Here is a good place to discuss the difference between the diffusion coefficient and the scattering coefficient:
 

4. How can scattering coefficients  be estimated?

This is likely to be the "FAQ of all FAQs" regarding software like CATT-Acoustic™ that handles diffusion and the answer is a combination of good and bad news. The bad news is that there exists no tables of data nearly as well spread and accepted as those for absorption coefficients. The data that does exist so far covers mainly commercially available diffusers and lists mainly the diffusion coefficient (see RPG site http://www.rpginc.com/research/index.htm and in the Product Specifications section, e.g. http://www.rpginc.com/products/qrd734/qrd_dc.htm and http://www.rpginc.com/products/skyline/sky_dc.htm), but the scattering coefficient for (say) a coffered ceiling or a statue will typically not be found, at least not in the near future. It is more complex to measure a diffuser than an absorber, at least more difficult than the random incidence absorption coefficient mostly used in RT calculations. One of the reasons is that the diffusion coefficient will vary with the size of the diffuser (a problem also present with absorbers but to a lesser degree) and another that a diffuser will typically not diffuse equally in all directions. The ISO project group [ISO-2000] has developed a measurement procedure to measure a random incidence scattering coefficient but regarding e.g. size dependence it has similar difficulties (perhaps especially so for practical reasons since the method involves rotating a sample). So, what can then possibly be the good news? For RT estimation purposes the scattering coefficients do not have to be very detailed and a bit of physical reasoning will in most cases suffice! For ordinary building elements the following scattering coefficients are recommended (note that CATT-Acoustic™ assumes coefficients to be entered as % values 0..100  rather that fractions 0..1)

• 0.10 - 0.20 minimum on all surfaces (0.08-0.10 on big flat ones)
• 0.40 - 0.70 for 125 Hz - 4 kHz on audience areas
• for generally rough surfaces set high values (0.80) where the roughness is of the order of or higher than the wavelength, and gradually lower below. For example, if the roughness scale is 0.3 m, set 0.80 for 1 - 4 kHz, 0.60 at 500 Hz, 0.30 at 250 Hz and 0.15 at 125 Hz.
• CATT-Acoustic™ has a feature called "Auto edge diffusion" that automatically calculates a size- and frequency-dependent edge diffusion so that smaller surfaces give more diffusion and weaker specular reflections at lower frequencies. This has successfully been applied to e.g. desks in classrooms and to reflectors, cupboards and similar that are hard and flat, and therefore can not be assigned significant surface diffusion. It can also be applied to patches of absorbers that have a diffusing effect. If a surface is both small and rough, it will have both types of diffusion.
• Generally, there is greater risk associated with underestimating scattering coefficients than with overestimating them.

It can be difficult to know in advance if a room is sensitive to diffusion settings or not (as indicated in Part I of the paper, it depends on the hall shape and the absorption distribution). A useful test is to calculate both with an initial reasonable guess of scattering coefficients and with diffusion switched off and see if the resulting RTs differ substantially. If so, the diffusion has to be more carefully estimated and it can be wise to include in the room design some options for final fine-tuning of the RT. This possibility to calculate both with and without diffusion is very useful. If a CP software does not handle diffuse reflection, there is no way of knowing how sensitive the RT prediction is and, as illustrated in Part I, the predicted RT can then be much too long (actually, with specular-only reflection it may also happen that the early RT is predicted too short while the late RT is predicted too long). Further, the predicted Sabine RT may be too short since that assumes fully diffuse reflections. It is physically impossible for a room to have all walls 100% specularly reflecting or all walls 100% diffusely reflecting. In the majority of cases rooms rather have varying degrees of partly diffusing walls and a CP program handling frequency dependent diffusion is required for the real-world cases in-between the two idealized extremes.

We now return to the sports hall from Part I where the acoustic consultant did a series of prediction tests to investigate what scattering coefficients would have given the measured T-30 in the initial case (5.7 sec) and the result was around 0.10 as is in accordance with the general recommendation of 0.10 above for large flat walls. Worth mentioning here is that the ceiling beams can be handled by auto edge diffusion. Table II-1 shows various predicted RT values for the 1 kHz octave-band where the Sabine and the specular-only CP cases form the low and high extremes while the CP values with diffusion taken into account are close to the measured one.
 

Table II-1 Measured and predicted RT values @ 1 kHz for the sports hall from Part I.

The sports hall also has another smaller, but still significant, estimation problem since nearly 90% of the surface in the detailed case is concrete. This means that the late decay (affecting T-30) is very dependent on the exact value of the absorption coefficient selected for concrete. Consider these two cases:

• a room with reasonable mixing or diffusion and varied absorption (concrete is say 10% of the total surface area): changing the absorption coefficient for concrete from 0.02 to 0.01 will hardly affect the RT at all 
• a room made only of concrete: changing the absorption coefficient from 0.02 to 0.01 will double the RT (except at higher frequencies where air absorption will dominate in a large room).

In other words, when a hard material is dominating a very accurate estimate of the absorption coefficient is necessary for purely numerical reasons. 

To conclude, this simple-looking sports hall actually demonstrated a range of problems regarding RT prediction and to arrive at 10% from the measured value is quite good, especially in relation to the very large errors given by the Sabine formula or a specular-only prediction. However, rooms that are more mixing and that have a more even absorption distribution, are not as sensitive to diffusion settings and tend to be considerably easier regarding prediction.

5. How will diffuse reflection affect auralization?

This short two-part paper about RT estimation and diffusion in CP programs has focused on what the author considers to be a first order effect of diffuse reflection when it comes to estimation of objective parameters such as the RT. However, with auralization also the subjective impression counts. There has not been many listening tests made regarding this but a fairly recent one involving CATT-Acoustic™ indicated that when scattering coefficients were varied listeners perceived the changes in an expected manner [Torres-00]. Basically, for the listener position studied, the article showed that:

1. frequency-dependent changes in the diffusion was audible.
2. the character of the perceived difference depended on the input signal (same impulse response and scattering coefficients but different perceived character). This is not surprising since if the input signal did not have significant energy in the frequency range where diffusion was varied not much difference could be perceived.
3. listeners gave entirely consistent answers in terms of the perceived frequency ranges where changes in the scattering coefficient occurred.

References, Part II

[AES-4id-2001] JAES 49(3), 149-165, 2001
[ISO-2000] ISO/CD 17497:2000, "Acoustics - Measurement of the random-incidence scattering coefficient of surfaces"
[Kuttruff] "Room Acoustics," H. Kuttruff, Applied Science Publishers Ltd. London
[Rindel-01] "Scattering in Room Acoustics and the Related Activities in ISO and AES," J. H. Rindel, 17th ICA Rome 2001, paper 6KN1.02
[Cox-01] "Contrasting surface diffusion and scattering coefficients", T. Cox, P. D'Antonio, 17th ICA Rome 2001, paper 6B.09.01
[Torres-00] "Audibility of 'Diffusion' in Room Acoustics Auralization: An Initial Investigation," R. Torres, M. Kleiner, B.-I. Dalenbäck, Acta Acustica 86(6), 917-925 (2000)

Home: Research & Development: Research Topics: Reverberation Time, Diffuse Reflection, Sabine, and Computerized Prediction - Part II