Enhanced Active Noise Control using Perceptual Filters and Bispectrum Energy

Abstract Active noise control (ANC) technology effectively reduces dynamic noise, and it is becoming increasingly common in wearable audio devices. The generative fixed-filter ANC (GFANC) approach creates optimal control filters from little prior data (i.e., a single pretrained broadband filter) by employing an adaptive subfilter combination. By using a preset set of pretrained filters, the GFANC improves adaptability to a variety of noise settings. The dynamic mixing of subcontrol filters enhances noise reduction; nevertheless, in situations where noise conditions change quickly, relying on current noise frames alone may result in inaccurate results.To improve system robustness, bispectrum energy (BE) analysis is a higher-order spectral technique used to capture nonlinear interactions and phase couplings among the frequency components of the noise signal. In contrast to conventional power spectrum techniques, BE offers more comprehensive statistical data regarding the noise structure, facilitating more precise identification and selection of subcontrol filters that more closely align with the intricate features of dynamic noise. This statistical insight helps mitigate errors caused by transient or nonstationary noise components.Noise weighting filters, such as A–weighting and ITU–weighting, are used to enhance perceptual quality by emphasizing the frequency ranges that are most sensitive to human hearing. This helps to more correctly model auditory perception. By prioritizing noise reduction in crucial frequency bands, these weighting filters increase the perceptual efficacy of noise suppression and enhance the wearable device's overall user experience.In terms of convergence speed and overall noise reduction effectiveness, numerical simulations using real-world noise recordings show that our suggested approaches perform noticeably better than conventional adaptive algorithms.