Fast learning for sound field interpolation using extreme learning machines for active noise control

Active noise control (ANC) generates anti-noise with the same amplitude and opposite phase as the target noise, thereby canceling it out. Performance is improved by utilizing the residual noise detected by error microphones placed at control points. In particular, multi-point ANC aims to control noise at multiple error microphone locations within a region of interest (ROI). Recently, to avoid physical constraints on user activity caused by placing microphones inside the ROI, methods using physics-informed neural networks (PINNs) have been proposed to interpolate the sound field within the ROI from a small number of microphones located outside it. However, these approaches require considerable learning time for the neural networks. In this study, we propose a sound field interpolation method for multi-point ANC based on the extreme learning machine (ELM), a single-layer feedforward neural network that enables fast learning. The proposed method learns efficiently from limited data and physical constraints to interpolate the primary sound field. Its effectiveness is demonstrated through simulation experiments. Work partially supported by Research Institute for Science and Technology of Tokyo Denki University Grant No. Q24J-04/Japan.