Hybrid Learning-based Active Noise Control in Encapsulated Structures

Aboutiman A, Karimi HR, Ripamonti F. Hybrid Learning-based Active Noise Control in Encapsulated Structures. In: 2025 AES International Conference on Artificial Intelligence and Machine Learning for Audio. Audio Engineering Society; 2025. Paper 5. Available from: https://aes.org/publications/elibrary-page/?id=22994

@inproceedings{Aboutiman2025_22994,
  author = {Aboutiman, Alkahf and Karimi, Hamid Reza and Ripamonti, Francesco},
  title = {{Hybrid Learning-based Active Noise Control in Encapsulated Structures}},
  booktitle = {2025 AES International Conference on Artificial Intelligence and Machine Learning for Audio},
  note = {Paper 5},
  year = {2025},
  month = sep,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=22994}
}

TY  - CPAPER
TI  - Hybrid Learning-based Active Noise Control in Encapsulated Structures
AU  - Aboutiman, Alkahf
AU  - Karimi, Hamid Reza
AU  - Ripamonti, Francesco
T2  - 2025 AES International Conference on Artificial Intelligence and Machine Learning for Audio
M1  - Paper 5
PY  - 2025
DA  - 2025/09/02
UR  - https://aes.org/publications/elibrary-page/?id=22994
PB  - Audio Engineering Society
LA  - en
AB  - Encapsulated environments present significant challenges for active noise control (ANC) due to their complex acoustic characteristics. A finite element model (FEM) was developed to characterize the vibroacoustic behavior of this encapsulated structure, which exhibits non-minimum phase characteristics. These non-minimum phase properties present additional challenges for noise control algorithms, as the systems response can initially amplify the noise before providing attenuation. Then, a hybrid ANC algorithm that combines deep learning with adaptive filtering. The proposed methodology involves both experimental and numerical modeling of the control environment, referred to as the "Noise Box," which replicates the acoustic conditions inside a vehicle. A finite element model (FEM) was developed to characterize the vibroacoustic behavior of this encapsulated structure, which exhibits non-minimum phase characteristics. These non-minimum phase properties present additional challenges for noise control algorithms, as the systems response can induce undershoot, phase shit or delay in the control. A two-dimensional convolutional neural network (2D CNN) to select the most suitable pre-trained control filter based on the primary noise characteristics. Experimental results demonstrate that the proposed learning-based ANC outperforms conventional ANC algorithms.
ER  -