Addressing aerodynamic noise in vehicles: a hybrid method for noise reduction and signal preservation

Abstract

Vehicle interior noise significantly affects passenger comfort and is composed of powertrain, road, and aerodynamic noise—the latter becoming dominant at high speeds. Advances in noise control and the rise of electric vehicles have diminished engine noise, thereby accentuating aerodynamic noise, which typically spans a frequency range of 40  Hz to 5  kHz. To address the challenges posed by this broadband, nonlinear noise, we propose a Hybrid Aerodynamic Active Noise Control (HAANC) methodology that integrates advanced signal processing and machine learning for precise noise suppression. Specifically, Variational Mode Decomposition (VMD) and the Hilbert-Huang Transform (HHT) are employed for high-resolution signal decomposition and feature extraction, while an adaptive multi-filter structure enhances computational efficiency and stability. In addition, a novel signal extraction network—built on convolutional and bidirectional long short-term memory layers with an attention mechanism—is designed to preserve in-vehicle target sounds, such as conversations and music, amidst interference. Experimental evaluations on real-world datasets show that for pure aerodynamic noise signals, Normalized Mean Squared Error (NMSE) decreased by approximately 10.23  dB and the power spectral density (PSD) was reduced by an average of 15.6  dB/Hz. Under mixed-signal conditions with a –10  dB signal-to-noise ratio, short-time objective intelligibility (STOI) improved by 6.2 %, Perceptual Evaluation of Speech Quality Mean Opinion Score (PESQ MOS) increased by 26.28 %, and Mean Opinion Score Listening Overall Opinion (MOS LOO) increased by 27.42 %. These quantitative improvements substantiate the superior performance and generalization capabilities of the proposed HAANC system for enhancing passenger comfort.