Contribution
Impact of Perceptual Error Shaping in Active Noise Control Across Signals with Diverse Spectral Tilt
* Presenting author
Abstract:
The objective of many traditional feedforward ANC systems, such as those based on the Filtered-Reference Least-Mean-Square (FxLMS) algorithm, is to minimize the squared error signal in order to achieve maximum mean noise reduction across the entire frequency spectrum. However, since human auditory perception is highly frequency-dependent, the noise reduction achieved by such systems may be suboptimal. To address this limitation, perceptual error shaping can be incorporated by applying a psychoacoustic rating filter, such as A-weighting, to the LMS input signals. In the present study, a perceptually motivated feedforward Filtered-Error LMS (FeLMS) ANC system is compared to a standard FxLMS implementation in a co-simulation, utilizing low-pass filtered and full-band realizations of white, pink, and brown noise, as well as a real-world recording of driving noise as innovation signals. The performance of both ANC implementations is evaluated in terms of the reduction of unrated and A-weighted noise levels after adaptation for an active headrest system and frontal and diffuse presentation of the innovation signals.