Comparative study on subband adaptive filtering algorithms for active broadband noise control in impulsive noise environment 2019-01-1528
Subband adaptive filtering (SAF) techniques have been increasingly used in active noise control, especially for acoustic broadband noise signal and system models with long impulse responses. In active noise control (ANC), the closed-loop delayless SAF schemes improve the convergence rate of the widely adopted conventional filtered-x LMS (FxLMS) algorithm in a more computationally efficient manner under wideband noise like colored signals or even nonstationary signals. In most real environment like vehicle interior cabin, however, the performance of ANC can be degraded by various outliers, including non-Gaussian impulsive signal such as transient impact acoustic response for road noise. Although several robust objective error criteria as well as threshold based LMS-type adaptive filtering algorithms have been investigated for active impulsive noise control, the computational burden and the requirement of a priori knowledge of noises would be hardly overcome for implementation or practically unavailable. In this paper, various state-of-the-art SAF algorithms with decorrelating property are studied in terms of the convergence property and computational complexity for broadband noise control in impulsive environment without requiring any prior information of noises since the subband decomposition might alleviate the sudden change of adaptive systems by spreading energy to multiple channels. The SAF algorithms with low-order norm of error evaluation and variable step sizes are further evaluated via simulations for the input of symmetric α-stable impulsive noise and colored noise in the impulsive environment, respectively. Results show that the delayless SAF algorithms appear more robust than the conventional LMS-type algorithms for impulsive noises. Moreover, the utilization of band-dependent variable step sizes for the delayless SAF algorithms significantly improves the convergence rate.
Guo Long, Kan Wang, Teik Lim
The University of Texas at Arlington
Noise and Vibration Conference & Exhibition