Project name: RT129 Author: Sam Blackburn Theatre: Maquette Zwolle Company: Adviesbureau Peutz Mook

Summarised report on the model measurements for the proposed theatre in Zwolle

Summary:
A new theatre has been planned in Zwolle. It is a complex design with variable acoustics and moveable parts, such as the ceiling panels, that can transform the space from a theatre into a concert hall, opera house or ballet venue: even musicals will be performed here, which wanting very different acoustics. Before it is built, Adviesbureau Peutz & Associates have been asked to make a model to check the proposed design for acoustic problems and to recommend optimum positions for the variables, such as ceiling positions and acoustic surfaces.



My work at Peutz was to prepare a system for acoustic testing, changing the model so that useful measurements could be taken, assemble the system and begin to take measurements. I spent 8 weeks working, starting with a bare MDF model and applying acoustic surfaces, fitting microphones and sources and turning it into a reliable system. I took measurements starting with an empty model and, taking a measurement after each change, gradually added parts to make a model with the acoustic properties of a concert hall full of people. I have also taken binaural measurements with a model of a dummy head, in order to make a recording of a piece of music “played” in the concert hall by the model source.

I have learnt a great deal about MLS measurements and put my understanding of linear systems into practice. I have discovered the problems with the mathematical techniques taught at university, such as noise “blowing up” an inverse FFT and the timing problems associated with filters.

Introduction:
MAXIMUM LENGTH SEQUENCES
Maximum Length Sequences can be used to find the impulse response of a system that is approximately linear. A certain binary sequence is fed through the system, producing many superimposed positive and negative step responses at the output, which can be cross-correlated with the input to find a single response. This method is more noise tolerant than taking the impulse response directly, because it puts more energy into the device under test than is possible with a single impulse without distorting or damaging the transducers.



In practice, this technique involves connecting a computer sound card to microphones and sources installed in the model (via amplifiers). Each part of this system must be checked for noise, distortion and filtering before the entire system can be trusted. If the result is being altered by the transfer function of another component, we need to compensate for it by calibrating the rest of the system.

TESTING AND CALIBRATION
To find the response of a model we must account for the response of the measuring system. Normally it might be reasonable to assume that measuring microphones and speakers have a flat frequency response, but this is unlikely when the microphones are being used above their nominal frequency range. A frequency response of the measuring equipment is found using a free field test, so that the results can be compensated afterwards to give just the characteristics of the model and not the measuring system. In theory, you can just “undo” the filtering by applying another filter which has the opposite effect on the frequency response.

Unfortunately there is a problem with this: the gain is very low indeed at some frequencies, and trying to put those frequencies back will amplify background noise by an intolerably large amount. We may end up with the whole impulse response disappearing behind noise at about 1Hz, even when this 1Hz noise was inaudible during the test.

But we are not interested in 1Hz anyway; we carefully need the measuring frequencies, so we can just filter out everything outside of the audible range and the noise will disappear. This would be excellent, but the filter has now changed the timing, so our echoes and decay times may inaccurate.

In practice, it is not possible to take a perfect impulse response. A good compromise is achieved with an uncorrected impulse response to analyse the timing, since this requires a minimum of filtering. If we want to listen to music in the model hall then we need a compensated frequency response, so we correct it with filters and do our best to remove the low frequency noise, although timing is always degraded in the process. The filtered response is convoluted with music played in an anechoic chamber, to make the sound that you should hear in a concert hall. Imperfections aside, we want to be sure that each component can perform well enough.

The Rest Of The Report
COMPANY SECRETS
Sorry, the rest of the report isn't online, because of course Peutz don't want their company secrets to be published! I hope that this preview gives a good impression of the project anyway. The following is explained in detail in the full report:

Microphones - the
requirements, and the
difficult job of mounting
them securely without
affecting the sound field.
Omnidirectionality and
frequency responses were
also checked.		 A page I prepared detailing symptoms, cause and solution to a problem experienced with the microphone cables. A plan I made of the model Zwolle concert hall showing source and microphone positions - click for larger image
Sources - a few
experiments to test
omnidirectionality,
and frequency response,
in a free field.
Positioning - placing
microphones in the
model to accurately
reflect the concert hall
as measured.		 A microphone on the left and conical source on the right, on the stage of the model.
Computers - running
standard software such as
MLSSA, as well as some
scripts written in house
by Peutz.		 The computer used to measure the model concert hall. A photomicrograph of MDF surface with various coatings of paint, which we analysed for porousity. Click for larger image.
Acoustically reflective
surfaces - photomicrographs
confirm that painted wood
is not porous, so will
reflect sound well.
Acoustically absorptive
surfaces - modelling
audience materal.		 A model reverberation room test, to find the absorption coefficient of foam.
An investigation into
absorption coefficients
for foam material.
Closing the model - the
use of absorptive material
e.g. cloth stuff, to reduce
leakage of noise into the
model.
Delay problems
Tests - Loopback etc.
Variables e.g. ceiling height
Variants: concert hall,
theatre, opera house.
Results e.g. RT60, clarity.
Conclusions
ACHIEVEMENTS
Most importantly, the Zwolle model is now ready for taking the rest of the measurements. Variants can be changed quickly, the measuring system is reliable, and the level of reproducibility is known. The first few measurements have been taken and I have shown my successor how to take some more. The results are organised and the report will be finished by the time you read this.

WORKING AT PEUTZ
The insight into MLS and the imperfect reality of nearly linear systems has been very interesting, as has the new experience of working in industry. I am glad that my boss did not mind my willingness to debate any points we disagree over, as this has sometimes been a point of tension at college. The only problem I had was being slow to adjust to the way people work: when I hit a problem I tended to wait until I can ask someone, instead of getting on with something else. This made me look like I wasn’t doing any work. Fortunately this problem went away once I had some idea of what I was doing.

WORKING IN MOOK
Mook is a very nice place to work, but accommodation was very hard to find. An employee let me use a spare room, but all other affordable places to stay were booked up by students at Nijmegen University long ago. Any future students should book their accommodation long in advance!

Top of Page