PhD, Finnur Pind - Wave Based Simulations & Virtual Acoustics

Finnur Pind defended his PhD at the Technical University of Denmark, the 31st of August, 2020. Finnur has contributed significant advancements in several interrelated fields such as wave based acoustic simulations and virtual acoustics through his research. The following is a superficial summary of Finnur’s research loosely based on, his published papers, his PhD defence, various presentations along with a short interview on his future plans. See references in chronological order in the bottom of this post.

Finnur presenting his research at Danish Sound Day 2019 where he won the prize for best research project.

Research background

Finnur often starts his presentations with a few facts on the state of room acoustics to provide the context for his work. Facts like, ”70% of people are unhappy with their sound environments in office buildings[1]”, “50 % of elementary schools don’t fulfil acoustics requirements[2]” or “0% of hospitals fulfil WHO’s noise criteria[3]” And when talking with Finnur, you get the clear sense, that the aim of his academic work has a healthy through line to the issues of real life and practical building design.

In collaboration with renowned architecture firm Henning Larsen Architects and Ecophon, Finnur covered a lot of ground through his research. From wave based 3d acoustic simulations, over mathematical computational optimization, to the use virtual acoustics in practical architectural design. The breath of his PhD is both deep and broad yet it has a clear focus – to improve the everyday sound environments of real people. His research can be seen as three punch combination: Improving acoustic simulation accuracy, improving computational efficiency of simulations and providing easily accessible design tools based on these simulations.

Finnur at Ecophon Central — Finnur wearing a 3d headset at Ecophon Central

Wave based simulation methods

Most room acoustic simulations anno 2020 is still made by so called geometrical methods. They are manageable for most specialised organisations in terms of processor power, but have several disadvantages in terms of accuracy. Namely, aspects of acoustic phenomena like diffraction, wave interference as well as scattering is not encompassed by these methods.

As an example the geometrical method of, “ray tracing” simulations entails having a simplified 3d modelled space where, as the name suggests, a multitude of rays (e.g. vectors) are traced and data obtained by interaction with the 3d space, constituting the simulation.

The simulation method of ray tracing was actually originally invented to simulate lightning – not sound.

In its original domain of simulation, the light waves are so small that their movement dynamics are quite similar to that of basic vector dynamics. Off course acoustic waves are many orders larger and actually does not have much dynamic overlap with vectors, especially in frequencies from around 500 Hz and below. Furthermore, most complexity of geometry is more or less left out of the simulations and so called “scattering coefficients” is assigned to simple surfaces as an approximation of the effects of more complex geometry.

In modern acoustic simulations, assigning scattering coefficients to a given surfaces or geometrical shape is often based on unsophisticated visual inspections or subjective “guestimations.”

It should be obvious by this cursory description that a lot of detail will be left out and/or misrepresented by this simulation method. Conversely, mapping wave dynamics on more detailed geometry will naturally yields far more precise results while also being far more computationally demanding.

Through three separate studies, Finnur provided advancements in wave based simulation in relation to two separate methods of computational processing: The spectral element Method (SEM) and High-Order Spectral Element Methods. In the case of the SEM, he presented a numerical scheme based on this method and assessed its suitability of room acoustic simulations. Computational efficiency was optimized by utilizing the so called implicit-explicit Runge-Kutta solver for time stepping. Results showed an impressive improvement in both “cost-efficiency” (processor power needed) and accuracy. In short this method is faster more precise while maintaining geometrical flexibility.

Impulse response vs. simulation — Frequency response simulation (coloured) and measurement results (dashed) of a cube shaped room showcasing the accuracy of Finnur’s method. Result pairs have been offset by 20 dB for better overview.

Following up on these results Finnur demonstrated how using high-order elements for capturing wave dispersion he was able to allow for a rougher 3d representation in the computation which again lowered processing power needed and therefore simulation time, while again maintaining a high accuracy especially in relation to curves, which is often a geometrical shape particularly hard to simulate with more conventional simulation methods.

Modelling boundary conditions

In addition Finnur also made advancements into modelling of boundary conditions, which simply speaking is the rules governing how simulated incident sound energy interacts with a given surface and what effects this has for the resulting reflected sound energy in the simulation. The accuracy of any acoustic simulation is highly dependent on the quality for the given boundary conditions.

Wave interaction with boundary conditions.

This represents a problem for acoustic simulations in general. As an example boundary conditions for typical room surfaces like porous materials installed with air cavity or hard backing was not well incorporated in simulation methods.

Finnur managed once again to present improvements in accuracy of modelling and uncertainty quantification of boundary conditions with little added computational requirements while integrating this knowledge into his existing methods as described above.

Virtual Acoustics

As covered thoroughly in other posts here on Acoustic Bulletin, acoustics have historically been systemically under prioritized in building design and the construction industry in general. The cause of this may very well be multi-facetted, but one hypothesis often referenced as significant is the elusive nature of acoustic data. For other professionals in the industry like architects, or laymen like clients or end users, it can be hard to conceptualise information in terms of numbers, graphs and or statistics in comparison to highly visual representations like plan drawings and renderings. Instead, acoustic data from a simulation can be converted in to an aural representation also called an auralization. When combined with virtual reality this method provides a more intuitive and holistic understanding of the effects of certain design decisions. Finnur often refer to this as “Virtual Acoustics” in his presentations. Virtual acoustics is off course significantly different from auralizations in particularly one aspect – Instead of simulating one fixed receiver position where the acoustics of a space can be experienced, in virtual acoustics the user should be able to move around freely in the entire space.

From an acoustic standpoint this provides a pedagogical 1-to-1 understanding of acoustic concepts that can be hard to grasp for layman with no experience in the field such as speech clarity or sound propagation. By moving around the virtual space the user can easily understand how acoustic parameters change with distance.

Virtual loudspeaker array, which follows the listener during run-time while interpolating impulse responses between presimulated points in the 3d model.

Virtual acoustics off course necessitates a far more comprehensive room acoustic simulation than a traditional auralization. In auralizations only one receiver point in the space is simulated whereas all points in a vertical plane has to be associated with an impulse response in virtual acoustics. In line with his other research, Finnur also made progress in computational efficiency in this domain. Roughly speaking, he accomplished this by presimulation and interpolation. By having a grid of receiver points presimulated roughly corresponding to a both computationally manageable and adequately detailed representation of the space he could calculate the intermediate receiver points with acceptable accuracy. In this way Finnur can present a highly realistic, impressively accurate and seamless virtual acoustic experience.

Finnur’s research further presented how his tools were used in practical design cases and Henning Larsen, an overview of the state of the science of virtual acoustics along with a discussion of the future of this technology.

Finnur is now moving back to his homeland of Iceland to continue his studies and work on his newly founded start-up “Treble” that will offer the high quality acoustics advice and simulations for the global construction industry.

Sources:

[1] A Briefing on Global Workplace Strategy, Management, Satisfaction & Effectiveness (2017)

[2] A survey of acoustic conditions and noise levels in secondary school classrooms in England (2014)

[3] Noise Levels in John Hopkins Hospital (2005)