Paper Review—Acoustic Voxels: Computational Optimization of Modular Acoustic Filters

Main Contribution

The research problem in this paper is computational design and fabrication in acoustic filters. Acoustic filters have many industrial and medical applications, but they are difficult to customize with desired properties, requiring heavy trial-and-error iterations. Driven by the effectiveness and efficiency of additive manufacturing in manufacturing complex shapes, the authors are motivated to utilize computational methods to simulate and optimize the shape of the cavity for the acoustic filtering effects, which are to be physically realized via additive manufacturing. This paper proposed Acoustic Voxels, as a computational method to assemble shape primitives into a complex geometry that produces the desired acoustic filtering.

The main contributions are that the proposed method largely automates the design of acoustic filters with custom properties; the range of acoustic filters is expanded and it opens up explorations into new possibilities such as mufflers design, wind instrument prototyping, and audiovisual UI via acoustic signatures.

Method

The authors first introduce background theory for acoustic filters, including how to evaluate acoustic filters - input impedance and transmission loss. However, optimizing a complex structure targeting either quantity is computationally expensive, thus in this project, the authors leverage the concept of the transmission matrix, which is widely used in industrial muffler design.

To use this technique in a filter structure, the authors assumed that acoustic pressure and velocity are both distributed uniformly over the cross-section, their relationship can be approximated linearly. Then, the authors go into details of the method design to construct the internal structure of an acoustic filter.

The method has roughly 3 steps. Firstly is to utilize a modular primitive resonator, which is a hollow cube with extruded cylinders on its six faces, where all the cylindrical extrusions have the same radius and length. This is advantageous because the primitive is small enough to fill any shape, flexible for sampling, and it leads to efficient computation of the transmission matrix. The second step is to assemble the primitives by connecting an inlet and an outlet to form a complex structure. The third step features an efficient combinatorial and continuous optimization algorithm of the assembly connectivity and individual primitive parameters. Random samples for the connectivity of the resonators are evaluated while continuously changing the size of each hollow cube. The authors then discussed 3 key applications of the system. This includes muffler design at a finer granularity thanks to the system’s ability to construct complex structures, for example, an engine noise muffler demonstrated in the supplementary video. Then, to present the system’s capability for prototyping an acoustic resonator which serves as the key part of wind instruments.

The authors present ways to design trumpets with customized sets of notes and shapes by maximizing the impedance values at the frequencies of those notes. The last application lies in acoustic signatures, such as acoustic tagging and encoding, where objects and devices (users) could interact through acoustics.

What do I think?

This paper spent more paragraphs explaining the key applications made possible by this novel system. I find it helpful in assisting people who don’t have much prior knowledge about sound simulation and acoustic filters (such as myself) to appreciate the value the proposed system brings forward.

In the supplementary video, the authors demonstrated how each application would work with a customized setting to achieve the desired acoustic properties. I especially find the last application particularly interesting in expanding new possibilities of audiovisual HCI design.

Considering acoustic encoding, the nature of encoding makes it a more secure and private way to access and store information. This reminds me of the privacy-preserving discussions in NLP - with the emergence of AI assistants (such as Alexa) and wearable technologies (smart watch etc.), the users could be vulnerable to data breaches and privacy infringements. I wonder if acoustic encoding could be used in NLP applications to preserve users’ privacy as I assume speech recognition is broken down into acoustic in the hardware. This may make an interesting future work branching out of the topic of this paper as well as research around NLP and personal assistance technologies.