Development and experimental verification of a new computationally efficient parallel narrowband active noise control system

The filtered-x least mean square based narrowband active noise control (FX-NANC) system with a parallel structure is typically used to cancel low-frequency multi-tonal noise. The computational efficiency and system performance may, however, be greatly degraded due to an increase in the number of controlled frequencies or the length of estimated secondary path. To alleviate this problem, a cost-saving FE-NANC system has been recently proposed, but the system suffers from slow convergence. This paper presents a local secondary path estimation based filtered-error least mean square (LFELMS) algorithm and applies it to a narrowband active noise control (NANC) system. By filtering a single error signal with a set of low-order local secondary path (LSP) models to obtain filtered signals for weight updating, the proposed LFELMS based NANC (LFE-NANC) system requires significantly less computational cost compared to the FX-NANC system and has considerably faster convergence speed than the FE-NANC system. Two new LSP modeling methods are also proposed. In addition, a computational complexity analysis of some typical NANC systems and the proposed LFE-NANC system is provided. Numerical simulations are performed to investigate the accuracy of the LSP modeling methods and compare the performance of the proposed system with other NANC systems. Finally, real-time experiments are conducted to confirm the effectiveness of the proposed system in single-tone, multi-tone and non-stationary situations.