Contribution
Pilot study on neural virtual sensing for active noise control
* Presenting author
Abstract:
The majority of local active noise control algorithms are based on adaptive filters, which require the residual error at the listening position. As microphones typically cannot be placed at the target position without disturbing the listener, the error signal can be estimated from nearby microphones using virtual sensing approaches. Classical virtual sensing methods are generally highly dependent on the specific acoustic environment and geometric arrangement. This implies that a variety filter sets must be calculated in advance and switched during operation, with each set optimized for a distinct acoustic scenario and position. This contribution presents a novel lightweight neural virtual sensing approach, based on a filter-and-sum beamformer. The impact of different input features on error signal estimation is investigated in a pilot study. This approach enables the handling of various acoustic scenarios with a single model, obviating the necessity for switching logic in time variant systems while maintaining computational resource efficiency for embedded systems.