Proof-of-Concept of Boundary Neural Net

Project title: Proof-of-Concept of Boundary Neural Net 


This project is a proof-of-concept study of a newly developed machine learning (ML) based inverse characterization of boundary conditions in 3D room called Boundary Neural Net (BNN). This particular method was developed over the last 6 months during Yuanxin Xia’s MSc project (who is one of the main researchers for the project applied).  

The current standardized methods for characterizing typical building materials’ acoustic properties are outdated and not accurate enough. For example, ISO 10534 is only a ‘normal incidence’ surface impedance/absorption coefficient measurement, while oblique incidence is likely to occur in 3D rooms. One needs to take a small sample (max 10 cm edge length) and fit it into a tube for controlled measurement procedures, and therefore full-scale characterization is prohibited. ISO 354, the reverberation chamber method, is extremely inaccurate and too many non-acoustic factors play roles during the measurement. Moreover, the results obtained according to ISO 354 are hardly useful in room acoustic simulation for two reasons:

1) the absorption coefficients are often higher than 1, meaning more than 100 % of the incident energy is absorbed which is non-physical.

2) The absorption properties vary with the size of the specimen, meaning the results are contaminated by non-acoustical factors. For those reasons, the characterization of absorbers is often regarded untrustworthy. All these factors call for a new, robust and reliable characterization method. 

We developed a data-driven method that relies on neural networks to infer and restore the acoustic properties of all the surfaces based on sound pressure measurements. So far only simulated measurements are used in 2D. The proposed method is physics-informed, meaning we use the wave equation or Helmholtz equation to improve the accuracy when the data is scarce. This is a novel development that can revolutionize the acoustic characteristics of absorbers and diffusers in future. Consequently, Ecophon and Alpha Akustik that are manufacturing / consulting customers will take part in this project as industrial partners. HBK, a manufacturer of sound measuring systems will help validate the method in actual rooms.


Start: 15 September 2023
End: 31 January 2024
Funding: 134.400 DKK

Partners Industry:

Saint-Gobain Ecophon 

Alpha Akustik

Hottinger, Brüel & Kjær


Ecophon RB
Alpha akustik rb canva

Lead Research:


Cheol-Ho Jeong, Department of Electrical and Photonics Engineering 

Danish Sound Cluster professional focus area:

Future Sound Tech Solutions


Funded by:

Uddannelses- og Forskningsministeriet (UFS)

Danish Agency for Higher Education and Science


UFS logo rb

Danish Sound Cluster contact person:

Project Manager Tinne Midtgaard, – tlf. 3049 7846

Danish Sound Cluster

Hold dig opdateret

Tilmeld vores nyhedsbrev: