AURA Simulation Parameters - what should I use when?
AURA is a highly sophisticated room-acoustic module. It has been integrated into EASE, based on the CAESAR software from Aachen University. It can be considered as consisting of two different engines. One method is utilized for highly accurate auralisations at selected receiver positions, it is called AURA Response. The other one is less precise but much faster and therefore allows for the calculation at many points, that means AURA Mapping.
In general for medium size rooms particle numbers of about 100 000 particles should be used. This particle number determines the rays to be sent out for the Mapping, and in case of the Response calculation, the rays to be sent out for the diffuse part and for the late specular part, after the cutoff order. For the AURA Response calculation also another particle number matters, the number of rays sent out to systematically scan (a.k.a. cone tracing) the room to find specular reflections by means of mirror image sources. It can be calculated from the receiver diameter and it is approximately 2 500 000 times the inverse of the squared diameter. Accordingly the diameter should be chosen in the order of modeling detail, typically between 0.5m and 10m. For normal purposes a diameter of 5m turns out to be sufficient.
The length of the particle simulation should depend on the estimated broadband reverberation time (RT), that means it should be at a minimum 2/3rds of this time.
Furthermore for rooms with strongly non-convex shapes, such as coupled volumes the diffuse rain method is recommendable. It will increase the amount of diffuse energy obtained at receiver locations, that are only seldom reached directly by a particle. Here the more systematic rain method helps to build up a good diffuse field. When using diffuse rain roughly 1/10th of the usual particle numbers is appropriate.
The density factor determines how many reflectogram pulses are generated in average per ms for the late part. In many cases a number around 20 is fine. Also, this does not increase the simulation time, as the pulse generation is a post-processing step. But, do not choose it to large, otherwise the reflectograms become very huge.
The tail resolution also applies to the resolution of the late part, namely how many ms-bins are used for the particle arrivals. This number typically increases the memory requirements, but not the calculation time.
The cutoff order sets the limit order for the specular raytracing part in AURA response, as described before. For very diffuse rooms, specular reflections may only be expected for the first few orders, so a very low number can be used. However, for very hard rooms, for example with flutter echos, a high number is needed to create a realistic listening impression.
The specular part dominates mainly the feeling for directions and time arrivals of reflections (and so for the location of walls), while the diffuse part adds more feeling for the "color" and for the spaciousness of the room. Depending on the practical circumstances the simulation of either one can be shortened by choosing appropriate parameters: Generally cutoff order and receiver diameter for the first, particle number and length for the second.
Finally for AURA mapping you can also choose between Energy Loss and Particle Loss. Particle Loss is fine if you are only interested in energy ratios and energy in intervals. Here the temporal fine structure of the reflectogram/echogram does not matter very much. However, for the calculation of STI and other metrics relying on the details of the impulse response, time bins with no or very few particles are bad. Here a good coverage with particles over time is needed. The calculation of the RT is located somewhere inbetween, often energy loss with low particle number already yields satisfying results. Always make sure you look at the echograms/histograms and check for numerical problems. Strange spikes or empty time intervals always indicate a particle number that is too low.
As a conclusion should be stated that the simulation has reached a good quality if the results do not change anymore when the resolution is increased. This is always the main point for any numerically obtained result!
If the calculation times are still to high, try to reduce the detail of your model. The experience of many people with room-acoustic modeling is that a number of few hundred faces for ANY space is sufficient.
There is no need to model the knob of a door, because it is neither acoustically relevant nor can you model the rest of the room with such accuracy. For example the acoustic data of wall materials and loudspeakers are 1/3rd octave. The dynamic properties of air due to thermal effects are not taken into account, as well as low-frequency modes and a moving audience.