To characterise an architectural space, we measure a set of impulse response. You could measure one of these by creating a short impulsive sound and picking up the sound elsewhere on a microphone. Fig. 2 is an example from the model scale measurement for the source near the altar stone and the microphone inside the outer sarsen stones. First the sound direct from source to receiver arrives, followed quickly by a series of reflections from various stones.
Impulsive sounds can be used, for example you can make quick room measurements by bursting a balloon. But more accurate results are obtained using a signal that sweeps through all the frequencies from low to high in a quick chirp (Fig. 3). A bit of mathematical processing then gives the impulse response.
For this 1:12 scale model, we shrunk all the dimensions of the structure by a factor of 12. This means testing in the model using sound at twelve times the frequency. The reason you increase the frequency is to make the relative size of the sound waves and stone dimensions the same in the model as it would be in the real full-scale stone circle. This is vital so the sound interacts with the model stones in a way that mimics the real site at full-scale.
The chirp signal went up to 96,000 Hz, which means in theory we might get data up to about 96,000/12 = 8,000 Hz full-scale. In reality, the sources and microphones limited that range further.
Switching from the model scale measurements to the full-scale results is easy. The measurements were taken with a sampling frequency of 192,000 Hz. When we analysed we just reduced the sampling frequency by a factor of 12 i.e. 192,000/12 = 16,000 Hz.
Note, all the graphs in this blog have the times and frequencies at the full-scale equivalents.
The 1/4″ measurement microphone we used can measure up to 100,000 Hz. The problem is that 1/4″ microphones and pre-amplifiers are noisy due to thermal and electrical noise. Broadband ultrasonic microphones with high sensitivity are not common.
This was particularly a problem for Stonehenge because it’s not an enclosed space. Consequently, a lot of the sound energy put into the model gets rapidly lost and absorbed by the walls of the semi-anechoic chamber. The way we solved this was to make measurements using 128 chirps and averaging the results to reduce the electrical and thermal noise.
Loudspeakers also pose a problem, because we require them to work over at least six octaves and be small enough to fit into the model. As we’re working beyond the audible frequency range, there are limited choices because not many people make loudspeakers that span the range we need. Bowers and Wilkins kindly gave us a diamond tweeter that we used to give the broadest bandwidth for auralisation. We’ve assumed that this has a directivity similar to a human talker, but we need to measure to check that.
Architectural acoustics is normally tested with loudspeakers that radiate evenly in all directions (‘omnidirectional’). Consequently we arranged four ring tweeters to achieve a more uniform radiation of sound horizontally (see Fig. 5 for the directivity). The source is shown in Fig. 4 near to the altar stone.
We also used a 3.5″ tweeter in a cabinet for low frequency measurement.
Fig. 6 shows the spectrum for the measurement vs the background noise (a measurement with the loudspeaker turned off). It shows the measurements most vulnerable to noise are at high frequency, because the impulse response naturally has less energy at high frequency. So below I just show the plots in the 4000 Hz band.
Architectural acoustic involves analysing the sound reflections as they die away. Fig 7. shows an example of this for the 4000 Hz octave band. Typically we’d want the measurement to start at least 45 dB above the background noise floor, and this is achieved in this case.
A key issue in scale model testing is to scale the acoustic properties. For example the stones should absorb the same amount of energy at 12,000 Hz in the model, as they would do at 1,000 Hz in the real stone circle.
An important source of sound absorption in models is air absorption. This is the loss of sound energy as the sound waves go through the air. This absorption is mostly not important in architectural acoustics, but it’s important in scale models because air absorption rapidly increases with frequency. For example the air absorption at 48,000 Hz in the model is 4 times greater than it would be at full scale 4,000 Hz.
To correct the impulse responses for air absorption in each octave band a correction curve was applied . This only makes a significant different in reverberation times at 2000 and 4000 Hz, increasing the decay times by about 0.1 and 0.2 seconds respectively for the measurement example shown in this blog.
Calculating reverberation time
To extract the reverberation time, we first need to calculate the Schroeder curve . Reverberation time is defined as the time it takes a steady-state sound to decay by 60dB once the source is turned off (the interrupted noise method). Our measurements use short impulsive sound, which is a bit different. The Schroeder curve converts the impulse measurements to the decay curve you’d get from interrupting steady-state noise.
An example is shown in Fig. 8 (details of this calculation are in the Appendix). The reverberation time is calculated by fitting a regression line from -5 to -35 dB on the Schroeder Curve . In this case, the reverberation time is 0.65 s (the time for sound to decay by 60 dB).
Any measurement details I’ve missed you’d like to know? Please comment below.Continue reading