A combined method for multi-channel active noise control based on online adaptive estimation and offline convolutional neural network

Abstract

Multi-channel active noise control (ANC) has been widely used to attenuate low-frequency noise in relatively large spatial areas. Traditional multi-channel systems are achieved by optimizing the controller weights with adaptive algorithms that minimize the power of all error signals. However, nonlinearity in ANC systems can significantly degrade the performance of conventional adaptive algorithms, which is even more severe in multi-channel systems. Researchers suggested using neural networks (NN) to deal with the nonlinear problems. However, applying NN to multi-channelANC systems is challenging in real-time processing because of the high computational burden. Therefore, a multi-channel ANC method that combines NN and adaptive method is proposed. The proposed controller consists of both linear and nonlinear parts, where the linear part is estimated adaptively to track the change of primary noise in real applications while the nonlinear part is modeled offline with a small-scale convolutional NN (CNN). This method has the advantages of solving the nonlinear problem in ANC systems and achieving the capability in tracking primary noise source positions. At last, the performance of the proposed algorithm is verified through simulation experiments.