Hybrid Learning-based Active Noise Control in Encapsulated Structures

Active noise control (ANC) in enclosed environments such as vehicle cabins presents substantial challenges due to complex acoustic paths. To address this, a hybrid ANC strategy is proposed, combining adaptive filtering with deep learning techniques. A finite element model (FEM) was developed to simulate the vibroacoustic dynamics of a closed test structure referred to as the “Noise Box” which reproduces the acoustic conditions inside a car. This model highlights the presence of non-minimum phase characteristics, which complicate control by introducing delays, undershoots, or initial noise amplification before attenuation occurs. To improve control performance under these conditions, a two-dimensional convolutional neural network (2D CNN) is used to select the most appropriate pre-trained control filter based on the spectral features of the incoming noise. Experimental results show that this learning-based hybrid ANC approach achieves superior noise reduction compared to conventional adaptive algorithms.