ELSTM-ANC-OSPM: Enhanced LSTM in Active Noise Control Systems With Online Secondary Path Modeling

Active noise control (ANC) technology based on the fixed-coefficient filter strategy can effectively attenuate Gaussian noise for ANC systems. However, the traditional selective filter ANC with a convolutional neural network (SFANC-CNN) is incapable of combating non-Gaussian noise. To solve this problem, an enhanced long short-term memory (LSTM) neural network with the online secondary path modeling for ANC is proposed, termed ELSTM-ANC-OSPM algorithm, which includes the pre-training stage and noise control stage, and can attenuate the α-stable noise, K-distributed noise, mixed-Gaussian noise, chaotic noise, and the Gaussian noise. The ELSTM-ANC-OSPM algorithm integrated with the filtered-x maximum correntropy entropy criterion (FxMCC), whose kernel bandwidth is determined by looping over 0.01 to 1 with the step size of 0.01, to pre-train the control filters. Particularly, a novel hybrid LSTM-CNN structure, which contains two-layer of LSTM and three-layer of CNN, is proposed for online secondary path modeling. Compared with the CNN-based ANC algorithm, the ELSTM-ANC-OSPM algorithm achieves improved noise reduction and reduced computational time.