Hybrid deep learning–based active noise control for encapsulated structures with openings

Traditional active noise control (ANC) systems rely on adaptive filtering techniques that assume a linear relationship between signals. However, in practical implementations, nonlinear distortions, particularly those introduced by loudspeakers, can degrade control performance, limiting the effectiveness of conventional ANC approaches. This paper presents a hybrid ANC algorithm, GFANC-THFxNLMS, designed to enhance adaptability and robustness in complex acoustic environments, such as encapsulated structures with openings. The proposed method leverages deep learning to construct a dataset of pre-trained sub-control filters and employs a 1D convolutional neural network (CNN) to dynamically generate the most suitable filter for the incoming noise. An adaptive filtering stage then fine-tunes the control filter in real time, ensuring optimal noise attenuation. The performance of GFANC-THFxNLMS is evaluated against conventional ANC algorithms, namely GFANC and THFxNLMS, under both time-invariant and time-varying conditions. The results indicate that the proposed method achieves enhanced noise reduction, with 6.74 dB using THFxNLMS, 8.01 dB using GFANC, and 9.49 dB using GFANC-THFxNLMS, highlighting its effectiveness in noise mitigation.