An Investigation of Active Noise Control Based on Wave-U-Net

With the advancement of modern society, residents’ demand for a healthier and more comfortable living environment is increasing, and the noise problem has become a focus of attention. Traditional active noise control (ANC) methods such as the Filtered-x Least Mean Square (FxLMS) algorithm have limited effectiveness in dealing with systems containing nonlinear distortions, which restricts their application in practice. In this paper, we describe ANC as a supervised learning problem to deal with nonlinear distortions and propose a method based on Wave-U-Net to build an ANC system. The core idea is to simulate the adaptive filter in the FxLMS algorithm. Four loudspeaker nonlinearity degrees are applied to train the network, and various noises under different nonlinearity degrees are employed to test against the proposed ANC algorithm. Experimental results show that, compared to the FxLMS algorithm, the proposed ANC method has a faster response speed to noise and exhibits a good noise reduction effect indicating that it can effectively handle the system’s nonlinearity. Moreover, the method performs well for wideband noise and is robust to noise variations.