Implementation of Multi-channel Active Noise Control based on Back-propagation Mechanism (2208.08086v1)

Abstract: Active noise control (ANC) systems can efficiently attenuate low-frequency noises by introducing anti-noises to combine with the unwanted noises. In ANC systems, the filtered-x least mean square (FxLMS) and filtered-X normalized least-mean-square (FxNLMS) algorithm are well-known algorithms for adaptively adjusting control filters. Multi-channel ANC systems are typically required to attenuate unwanted noises in a large space. However, open-source implementations of the multi-channel FxLMS (McFxLMS) and multi-channel FxNLMS (McFxNLMS) algorithm continue to be scarce. Therefore, this paper proposes a simple and effective implementation approach of the McFxLMS and McFxNLMS algorithm. Motivated by the back-propagation process during neural network training, the McFxLMS and McFxNLMS algorithm can be implemented via automatic derivation mechanism. We implemented the two algorithms using the automatic derivation mechanism in PyTorch and made the source code available on GitHub. This implementation method can improve the practicality of multi-channel ANC systems, which is expected to be widely used in ANC applications.